Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * mm/kmemleak.c
4 *
5 * Copyright (C) 2008 ARM Limited
6 * Written by Catalin Marinas <catalin.marinas@arm.com>
7 *
8 * For more information on the algorithm and kmemleak usage, please see
9 * Documentation/dev-tools/kmemleak.rst.
10 *
11 * Notes on locking
12 * ----------------
13 *
14 * The following locks and mutexes are used by kmemleak:
15 *
16 * - kmemleak_lock (rwlock): protects the object_list modifications and
17 * accesses to the object_tree_root. The object_list is the main list
18 * holding the metadata (struct kmemleak_object) for the allocated memory
19 * blocks. The object_tree_root is a red black tree used to look-up
20 * metadata based on a pointer to the corresponding memory block. The
21 * kmemleak_object structures are added to the object_list and
22 * object_tree_root in the create_object() function called from the
23 * kmemleak_alloc() callback and removed in delete_object() called from the
24 * kmemleak_free() callback
25 * - kmemleak_object.lock (spinlock): protects a kmemleak_object. Accesses to
26 * the metadata (e.g. count) are protected by this lock. Note that some
27 * members of this structure may be protected by other means (atomic or
28 * kmemleak_lock). This lock is also held when scanning the corresponding
29 * memory block to avoid the kernel freeing it via the kmemleak_free()
30 * callback. This is less heavyweight than holding a global lock like
31 * kmemleak_lock during scanning
32 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
33 * unreferenced objects at a time. The gray_list contains the objects which
34 * are already referenced or marked as false positives and need to be
35 * scanned. This list is only modified during a scanning episode when the
36 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
37 * Note that the kmemleak_object.use_count is incremented when an object is
38 * added to the gray_list and therefore cannot be freed. This mutex also
39 * prevents multiple users of the "kmemleak" debugfs file together with
40 * modifications to the memory scanning parameters including the scan_thread
41 * pointer
42 *
43 * Locks and mutexes are acquired/nested in the following order:
44 *
45 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
46 *
47 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
48 * regions.
49 *
50 * The kmemleak_object structures have a use_count incremented or decremented
51 * using the get_object()/put_object() functions. When the use_count becomes
52 * 0, this count can no longer be incremented and put_object() schedules the
53 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
54 * function must be protected by rcu_read_lock() to avoid accessing a freed
55 * structure.
56 */
57
58#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
59
60#include <linux/init.h>
61#include <linux/kernel.h>
62#include <linux/list.h>
63#include <linux/sched/signal.h>
64#include <linux/sched/task.h>
65#include <linux/sched/task_stack.h>
66#include <linux/jiffies.h>
67#include <linux/delay.h>
68#include <linux/export.h>
69#include <linux/kthread.h>
70#include <linux/rbtree.h>
71#include <linux/fs.h>
72#include <linux/debugfs.h>
73#include <linux/seq_file.h>
74#include <linux/cpumask.h>
75#include <linux/spinlock.h>
76#include <linux/module.h>
77#include <linux/mutex.h>
78#include <linux/rcupdate.h>
79#include <linux/stacktrace.h>
80#include <linux/cache.h>
81#include <linux/percpu.h>
82#include <linux/memblock.h>
83#include <linux/pfn.h>
84#include <linux/mmzone.h>
85#include <linux/slab.h>
86#include <linux/thread_info.h>
87#include <linux/err.h>
88#include <linux/uaccess.h>
89#include <linux/string.h>
90#include <linux/nodemask.h>
91#include <linux/mm.h>
92#include <linux/workqueue.h>
93#include <linux/crc32.h>
94
95#include <asm/sections.h>
96#include <asm/processor.h>
97#include <linux/atomic.h>
98
99#include <linux/kasan.h>
100#include <linux/kmemleak.h>
101#include <linux/memory_hotplug.h>
102
103/*
104 * Kmemleak configuration and common defines.
105 */
106#define MAX_TRACE 16 /* stack trace length */
107#define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
108#define SECS_FIRST_SCAN 60 /* delay before the first scan */
109#define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
110#define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
111
112#define BYTES_PER_POINTER sizeof(void *)
113
114/* GFP bitmask for kmemleak internal allocations */
115#define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
116 __GFP_NORETRY | __GFP_NOMEMALLOC | \
117 __GFP_NOWARN)
118
119/* scanning area inside a memory block */
120struct kmemleak_scan_area {
121 struct hlist_node node;
122 unsigned long start;
123 size_t size;
124};
125
126#define KMEMLEAK_GREY 0
127#define KMEMLEAK_BLACK -1
128
129/*
130 * Structure holding the metadata for each allocated memory block.
131 * Modifications to such objects should be made while holding the
132 * object->lock. Insertions or deletions from object_list, gray_list or
133 * rb_node are already protected by the corresponding locks or mutex (see
134 * the notes on locking above). These objects are reference-counted
135 * (use_count) and freed using the RCU mechanism.
136 */
137struct kmemleak_object {
138 spinlock_t lock;
139 unsigned int flags; /* object status flags */
140 struct list_head object_list;
141 struct list_head gray_list;
142 struct rb_node rb_node;
143 struct rcu_head rcu; /* object_list lockless traversal */
144 /* object usage count; object freed when use_count == 0 */
145 atomic_t use_count;
146 unsigned long pointer;
147 size_t size;
148 /* pass surplus references to this pointer */
149 unsigned long excess_ref;
150 /* minimum number of a pointers found before it is considered leak */
151 int min_count;
152 /* the total number of pointers found pointing to this object */
153 int count;
154 /* checksum for detecting modified objects */
155 u32 checksum;
156 /* memory ranges to be scanned inside an object (empty for all) */
157 struct hlist_head area_list;
158 unsigned long trace[MAX_TRACE];
159 unsigned int trace_len;
160 unsigned long jiffies; /* creation timestamp */
161 pid_t pid; /* pid of the current task */
162 char comm[TASK_COMM_LEN]; /* executable name */
163};
164
165/* flag representing the memory block allocation status */
166#define OBJECT_ALLOCATED (1 << 0)
167/* flag set after the first reporting of an unreference object */
168#define OBJECT_REPORTED (1 << 1)
169/* flag set to not scan the object */
170#define OBJECT_NO_SCAN (1 << 2)
171/* flag set to fully scan the object when scan_area allocation failed */
172#define OBJECT_FULL_SCAN (1 << 3)
173
174#define HEX_PREFIX " "
175/* number of bytes to print per line; must be 16 or 32 */
176#define HEX_ROW_SIZE 16
177/* number of bytes to print at a time (1, 2, 4, 8) */
178#define HEX_GROUP_SIZE 1
179/* include ASCII after the hex output */
180#define HEX_ASCII 1
181/* max number of lines to be printed */
182#define HEX_MAX_LINES 2
183
184/* the list of all allocated objects */
185static LIST_HEAD(object_list);
186/* the list of gray-colored objects (see color_gray comment below) */
187static LIST_HEAD(gray_list);
188/* memory pool allocation */
189static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE];
190static int mem_pool_free_count = ARRAY_SIZE(mem_pool);
191static LIST_HEAD(mem_pool_free_list);
192/* search tree for object boundaries */
193static struct rb_root object_tree_root = RB_ROOT;
194/* rw_lock protecting the access to object_list and object_tree_root */
195static DEFINE_RWLOCK(kmemleak_lock);
196
197/* allocation caches for kmemleak internal data */
198static struct kmem_cache *object_cache;
199static struct kmem_cache *scan_area_cache;
200
201/* set if tracing memory operations is enabled */
202static int kmemleak_enabled = 1;
203/* same as above but only for the kmemleak_free() callback */
204static int kmemleak_free_enabled = 1;
205/* set in the late_initcall if there were no errors */
206static int kmemleak_initialized;
207/* set if a kmemleak warning was issued */
208static int kmemleak_warning;
209/* set if a fatal kmemleak error has occurred */
210static int kmemleak_error;
211
212/* minimum and maximum address that may be valid pointers */
213static unsigned long min_addr = ULONG_MAX;
214static unsigned long max_addr;
215
216static struct task_struct *scan_thread;
217/* used to avoid reporting of recently allocated objects */
218static unsigned long jiffies_min_age;
219static unsigned long jiffies_last_scan;
220/* delay between automatic memory scannings */
221static signed long jiffies_scan_wait;
222/* enables or disables the task stacks scanning */
223static int kmemleak_stack_scan = 1;
224/* protects the memory scanning, parameters and debug/kmemleak file access */
225static DEFINE_MUTEX(scan_mutex);
226/* setting kmemleak=on, will set this var, skipping the disable */
227static int kmemleak_skip_disable;
228/* If there are leaks that can be reported */
229static bool kmemleak_found_leaks;
230
231static bool kmemleak_verbose;
232module_param_named(verbose, kmemleak_verbose, bool, 0600);
233
234static void kmemleak_disable(void);
235
236/*
237 * Print a warning and dump the stack trace.
238 */
239#define kmemleak_warn(x...) do { \
240 pr_warn(x); \
241 dump_stack(); \
242 kmemleak_warning = 1; \
243} while (0)
244
245/*
246 * Macro invoked when a serious kmemleak condition occurred and cannot be
247 * recovered from. Kmemleak will be disabled and further allocation/freeing
248 * tracing no longer available.
249 */
250#define kmemleak_stop(x...) do { \
251 kmemleak_warn(x); \
252 kmemleak_disable(); \
253} while (0)
254
255#define warn_or_seq_printf(seq, fmt, ...) do { \
256 if (seq) \
257 seq_printf(seq, fmt, ##__VA_ARGS__); \
258 else \
259 pr_warn(fmt, ##__VA_ARGS__); \
260} while (0)
261
262static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
263 int rowsize, int groupsize, const void *buf,
264 size_t len, bool ascii)
265{
266 if (seq)
267 seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
268 buf, len, ascii);
269 else
270 print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
271 rowsize, groupsize, buf, len, ascii);
272}
273
274/*
275 * Printing of the objects hex dump to the seq file. The number of lines to be
276 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
277 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
278 * with the object->lock held.
279 */
280static void hex_dump_object(struct seq_file *seq,
281 struct kmemleak_object *object)
282{
283 const u8 *ptr = (const u8 *)object->pointer;
284 size_t len;
285
286 /* limit the number of lines to HEX_MAX_LINES */
287 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
288
289 warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len);
290 kasan_disable_current();
291 warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
292 HEX_GROUP_SIZE, ptr, len, HEX_ASCII);
293 kasan_enable_current();
294}
295
296/*
297 * Object colors, encoded with count and min_count:
298 * - white - orphan object, not enough references to it (count < min_count)
299 * - gray - not orphan, not marked as false positive (min_count == 0) or
300 * sufficient references to it (count >= min_count)
301 * - black - ignore, it doesn't contain references (e.g. text section)
302 * (min_count == -1). No function defined for this color.
303 * Newly created objects don't have any color assigned (object->count == -1)
304 * before the next memory scan when they become white.
305 */
306static bool color_white(const struct kmemleak_object *object)
307{
308 return object->count != KMEMLEAK_BLACK &&
309 object->count < object->min_count;
310}
311
312static bool color_gray(const struct kmemleak_object *object)
313{
314 return object->min_count != KMEMLEAK_BLACK &&
315 object->count >= object->min_count;
316}
317
318/*
319 * Objects are considered unreferenced only if their color is white, they have
320 * not be deleted and have a minimum age to avoid false positives caused by
321 * pointers temporarily stored in CPU registers.
322 */
323static bool unreferenced_object(struct kmemleak_object *object)
324{
325 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
326 time_before_eq(object->jiffies + jiffies_min_age,
327 jiffies_last_scan);
328}
329
330/*
331 * Printing of the unreferenced objects information to the seq file. The
332 * print_unreferenced function must be called with the object->lock held.
333 */
334static void print_unreferenced(struct seq_file *seq,
335 struct kmemleak_object *object)
336{
337 int i;
338 unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
339
340 warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
341 object->pointer, object->size);
342 warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
343 object->comm, object->pid, object->jiffies,
344 msecs_age / 1000, msecs_age % 1000);
345 hex_dump_object(seq, object);
346 warn_or_seq_printf(seq, " backtrace:\n");
347
348 for (i = 0; i < object->trace_len; i++) {
349 void *ptr = (void *)object->trace[i];
350 warn_or_seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
351 }
352}
353
354/*
355 * Print the kmemleak_object information. This function is used mainly for
356 * debugging special cases when kmemleak operations. It must be called with
357 * the object->lock held.
358 */
359static void dump_object_info(struct kmemleak_object *object)
360{
361 pr_notice("Object 0x%08lx (size %zu):\n",
362 object->pointer, object->size);
363 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
364 object->comm, object->pid, object->jiffies);
365 pr_notice(" min_count = %d\n", object->min_count);
366 pr_notice(" count = %d\n", object->count);
367 pr_notice(" flags = 0x%x\n", object->flags);
368 pr_notice(" checksum = %u\n", object->checksum);
369 pr_notice(" backtrace:\n");
370 stack_trace_print(object->trace, object->trace_len, 4);
371}
372
373/*
374 * Look-up a memory block metadata (kmemleak_object) in the object search
375 * tree based on a pointer value. If alias is 0, only values pointing to the
376 * beginning of the memory block are allowed. The kmemleak_lock must be held
377 * when calling this function.
378 */
379static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
380{
381 struct rb_node *rb = object_tree_root.rb_node;
382
383 while (rb) {
384 struct kmemleak_object *object =
385 rb_entry(rb, struct kmemleak_object, rb_node);
386 if (ptr < object->pointer)
387 rb = object->rb_node.rb_left;
388 else if (object->pointer + object->size <= ptr)
389 rb = object->rb_node.rb_right;
390 else if (object->pointer == ptr || alias)
391 return object;
392 else {
393 kmemleak_warn("Found object by alias at 0x%08lx\n",
394 ptr);
395 dump_object_info(object);
396 break;
397 }
398 }
399 return NULL;
400}
401
402/*
403 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
404 * that once an object's use_count reached 0, the RCU freeing was already
405 * registered and the object should no longer be used. This function must be
406 * called under the protection of rcu_read_lock().
407 */
408static int get_object(struct kmemleak_object *object)
409{
410 return atomic_inc_not_zero(&object->use_count);
411}
412
413/*
414 * Memory pool allocation and freeing. kmemleak_lock must not be held.
415 */
416static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
417{
418 unsigned long flags;
419 struct kmemleak_object *object;
420
421 /* try the slab allocator first */
422 if (object_cache) {
423 object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
424 if (object)
425 return object;
426 }
427
428 /* slab allocation failed, try the memory pool */
429 write_lock_irqsave(&kmemleak_lock, flags);
430 object = list_first_entry_or_null(&mem_pool_free_list,
431 typeof(*object), object_list);
432 if (object)
433 list_del(&object->object_list);
434 else if (mem_pool_free_count)
435 object = &mem_pool[--mem_pool_free_count];
436 else
437 pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
438 write_unlock_irqrestore(&kmemleak_lock, flags);
439
440 return object;
441}
442
443/*
444 * Return the object to either the slab allocator or the memory pool.
445 */
446static void mem_pool_free(struct kmemleak_object *object)
447{
448 unsigned long flags;
449
450 if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) {
451 kmem_cache_free(object_cache, object);
452 return;
453 }
454
455 /* add the object to the memory pool free list */
456 write_lock_irqsave(&kmemleak_lock, flags);
457 list_add(&object->object_list, &mem_pool_free_list);
458 write_unlock_irqrestore(&kmemleak_lock, flags);
459}
460
461/*
462 * RCU callback to free a kmemleak_object.
463 */
464static void free_object_rcu(struct rcu_head *rcu)
465{
466 struct hlist_node *tmp;
467 struct kmemleak_scan_area *area;
468 struct kmemleak_object *object =
469 container_of(rcu, struct kmemleak_object, rcu);
470
471 /*
472 * Once use_count is 0 (guaranteed by put_object), there is no other
473 * code accessing this object, hence no need for locking.
474 */
475 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
476 hlist_del(&area->node);
477 kmem_cache_free(scan_area_cache, area);
478 }
479 mem_pool_free(object);
480}
481
482/*
483 * Decrement the object use_count. Once the count is 0, free the object using
484 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
485 * delete_object() path, the delayed RCU freeing ensures that there is no
486 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
487 * is also possible.
488 */
489static void put_object(struct kmemleak_object *object)
490{
491 if (!atomic_dec_and_test(&object->use_count))
492 return;
493
494 /* should only get here after delete_object was called */
495 WARN_ON(object->flags & OBJECT_ALLOCATED);
496
497 /*
498 * It may be too early for the RCU callbacks, however, there is no
499 * concurrent object_list traversal when !object_cache and all objects
500 * came from the memory pool. Free the object directly.
501 */
502 if (object_cache)
503 call_rcu(&object->rcu, free_object_rcu);
504 else
505 free_object_rcu(&object->rcu);
506}
507
508/*
509 * Look up an object in the object search tree and increase its use_count.
510 */
511static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
512{
513 unsigned long flags;
514 struct kmemleak_object *object;
515
516 rcu_read_lock();
517 read_lock_irqsave(&kmemleak_lock, flags);
518 object = lookup_object(ptr, alias);
519 read_unlock_irqrestore(&kmemleak_lock, flags);
520
521 /* check whether the object is still available */
522 if (object && !get_object(object))
523 object = NULL;
524 rcu_read_unlock();
525
526 return object;
527}
528
529/*
530 * Remove an object from the object_tree_root and object_list. Must be called
531 * with the kmemleak_lock held _if_ kmemleak is still enabled.
532 */
533static void __remove_object(struct kmemleak_object *object)
534{
535 rb_erase(&object->rb_node, &object_tree_root);
536 list_del_rcu(&object->object_list);
537}
538
539/*
540 * Look up an object in the object search tree and remove it from both
541 * object_tree_root and object_list. The returned object's use_count should be
542 * at least 1, as initially set by create_object().
543 */
544static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias)
545{
546 unsigned long flags;
547 struct kmemleak_object *object;
548
549 write_lock_irqsave(&kmemleak_lock, flags);
550 object = lookup_object(ptr, alias);
551 if (object)
552 __remove_object(object);
553 write_unlock_irqrestore(&kmemleak_lock, flags);
554
555 return object;
556}
557
558/*
559 * Save stack trace to the given array of MAX_TRACE size.
560 */
561static int __save_stack_trace(unsigned long *trace)
562{
563 return stack_trace_save(trace, MAX_TRACE, 2);
564}
565
566/*
567 * Create the metadata (struct kmemleak_object) corresponding to an allocated
568 * memory block and add it to the object_list and object_tree_root.
569 */
570static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
571 int min_count, gfp_t gfp)
572{
573 unsigned long flags;
574 struct kmemleak_object *object, *parent;
575 struct rb_node **link, *rb_parent;
576 unsigned long untagged_ptr;
577
578 object = mem_pool_alloc(gfp);
579 if (!object) {
580 pr_warn("Cannot allocate a kmemleak_object structure\n");
581 kmemleak_disable();
582 return NULL;
583 }
584
585 INIT_LIST_HEAD(&object->object_list);
586 INIT_LIST_HEAD(&object->gray_list);
587 INIT_HLIST_HEAD(&object->area_list);
588 spin_lock_init(&object->lock);
589 atomic_set(&object->use_count, 1);
590 object->flags = OBJECT_ALLOCATED;
591 object->pointer = ptr;
592 object->size = size;
593 object->excess_ref = 0;
594 object->min_count = min_count;
595 object->count = 0; /* white color initially */
596 object->jiffies = jiffies;
597 object->checksum = 0;
598
599 /* task information */
600 if (in_irq()) {
601 object->pid = 0;
602 strncpy(object->comm, "hardirq", sizeof(object->comm));
603 } else if (in_serving_softirq()) {
604 object->pid = 0;
605 strncpy(object->comm, "softirq", sizeof(object->comm));
606 } else {
607 object->pid = current->pid;
608 /*
609 * There is a small chance of a race with set_task_comm(),
610 * however using get_task_comm() here may cause locking
611 * dependency issues with current->alloc_lock. In the worst
612 * case, the command line is not correct.
613 */
614 strncpy(object->comm, current->comm, sizeof(object->comm));
615 }
616
617 /* kernel backtrace */
618 object->trace_len = __save_stack_trace(object->trace);
619
620 write_lock_irqsave(&kmemleak_lock, flags);
621
622 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
623 min_addr = min(min_addr, untagged_ptr);
624 max_addr = max(max_addr, untagged_ptr + size);
625 link = &object_tree_root.rb_node;
626 rb_parent = NULL;
627 while (*link) {
628 rb_parent = *link;
629 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
630 if (ptr + size <= parent->pointer)
631 link = &parent->rb_node.rb_left;
632 else if (parent->pointer + parent->size <= ptr)
633 link = &parent->rb_node.rb_right;
634 else {
635 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
636 ptr);
637 /*
638 * No need for parent->lock here since "parent" cannot
639 * be freed while the kmemleak_lock is held.
640 */
641 dump_object_info(parent);
642 kmem_cache_free(object_cache, object);
643 object = NULL;
644 goto out;
645 }
646 }
647 rb_link_node(&object->rb_node, rb_parent, link);
648 rb_insert_color(&object->rb_node, &object_tree_root);
649
650 list_add_tail_rcu(&object->object_list, &object_list);
651out:
652 write_unlock_irqrestore(&kmemleak_lock, flags);
653 return object;
654}
655
656/*
657 * Mark the object as not allocated and schedule RCU freeing via put_object().
658 */
659static void __delete_object(struct kmemleak_object *object)
660{
661 unsigned long flags;
662
663 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
664 WARN_ON(atomic_read(&object->use_count) < 1);
665
666 /*
667 * Locking here also ensures that the corresponding memory block
668 * cannot be freed when it is being scanned.
669 */
670 spin_lock_irqsave(&object->lock, flags);
671 object->flags &= ~OBJECT_ALLOCATED;
672 spin_unlock_irqrestore(&object->lock, flags);
673 put_object(object);
674}
675
676/*
677 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
678 * delete it.
679 */
680static void delete_object_full(unsigned long ptr)
681{
682 struct kmemleak_object *object;
683
684 object = find_and_remove_object(ptr, 0);
685 if (!object) {
686#ifdef DEBUG
687 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
688 ptr);
689#endif
690 return;
691 }
692 __delete_object(object);
693}
694
695/*
696 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
697 * delete it. If the memory block is partially freed, the function may create
698 * additional metadata for the remaining parts of the block.
699 */
700static void delete_object_part(unsigned long ptr, size_t size)
701{
702 struct kmemleak_object *object;
703 unsigned long start, end;
704
705 object = find_and_remove_object(ptr, 1);
706 if (!object) {
707#ifdef DEBUG
708 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
709 ptr, size);
710#endif
711 return;
712 }
713
714 /*
715 * Create one or two objects that may result from the memory block
716 * split. Note that partial freeing is only done by free_bootmem() and
717 * this happens before kmemleak_init() is called.
718 */
719 start = object->pointer;
720 end = object->pointer + object->size;
721 if (ptr > start)
722 create_object(start, ptr - start, object->min_count,
723 GFP_KERNEL);
724 if (ptr + size < end)
725 create_object(ptr + size, end - ptr - size, object->min_count,
726 GFP_KERNEL);
727
728 __delete_object(object);
729}
730
731static void __paint_it(struct kmemleak_object *object, int color)
732{
733 object->min_count = color;
734 if (color == KMEMLEAK_BLACK)
735 object->flags |= OBJECT_NO_SCAN;
736}
737
738static void paint_it(struct kmemleak_object *object, int color)
739{
740 unsigned long flags;
741
742 spin_lock_irqsave(&object->lock, flags);
743 __paint_it(object, color);
744 spin_unlock_irqrestore(&object->lock, flags);
745}
746
747static void paint_ptr(unsigned long ptr, int color)
748{
749 struct kmemleak_object *object;
750
751 object = find_and_get_object(ptr, 0);
752 if (!object) {
753 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
754 ptr,
755 (color == KMEMLEAK_GREY) ? "Grey" :
756 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
757 return;
758 }
759 paint_it(object, color);
760 put_object(object);
761}
762
763/*
764 * Mark an object permanently as gray-colored so that it can no longer be
765 * reported as a leak. This is used in general to mark a false positive.
766 */
767static void make_gray_object(unsigned long ptr)
768{
769 paint_ptr(ptr, KMEMLEAK_GREY);
770}
771
772/*
773 * Mark the object as black-colored so that it is ignored from scans and
774 * reporting.
775 */
776static void make_black_object(unsigned long ptr)
777{
778 paint_ptr(ptr, KMEMLEAK_BLACK);
779}
780
781/*
782 * Add a scanning area to the object. If at least one such area is added,
783 * kmemleak will only scan these ranges rather than the whole memory block.
784 */
785static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
786{
787 unsigned long flags;
788 struct kmemleak_object *object;
789 struct kmemleak_scan_area *area = NULL;
790
791 object = find_and_get_object(ptr, 1);
792 if (!object) {
793 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
794 ptr);
795 return;
796 }
797
798 if (scan_area_cache)
799 area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
800
801 spin_lock_irqsave(&object->lock, flags);
802 if (!area) {
803 pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
804 /* mark the object for full scan to avoid false positives */
805 object->flags |= OBJECT_FULL_SCAN;
806 goto out_unlock;
807 }
808 if (size == SIZE_MAX) {
809 size = object->pointer + object->size - ptr;
810 } else if (ptr + size > object->pointer + object->size) {
811 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
812 dump_object_info(object);
813 kmem_cache_free(scan_area_cache, area);
814 goto out_unlock;
815 }
816
817 INIT_HLIST_NODE(&area->node);
818 area->start = ptr;
819 area->size = size;
820
821 hlist_add_head(&area->node, &object->area_list);
822out_unlock:
823 spin_unlock_irqrestore(&object->lock, flags);
824 put_object(object);
825}
826
827/*
828 * Any surplus references (object already gray) to 'ptr' are passed to
829 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
830 * vm_struct may be used as an alternative reference to the vmalloc'ed object
831 * (see free_thread_stack()).
832 */
833static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
834{
835 unsigned long flags;
836 struct kmemleak_object *object;
837
838 object = find_and_get_object(ptr, 0);
839 if (!object) {
840 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
841 ptr);
842 return;
843 }
844
845 spin_lock_irqsave(&object->lock, flags);
846 object->excess_ref = excess_ref;
847 spin_unlock_irqrestore(&object->lock, flags);
848 put_object(object);
849}
850
851/*
852 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
853 * pointer. Such object will not be scanned by kmemleak but references to it
854 * are searched.
855 */
856static void object_no_scan(unsigned long ptr)
857{
858 unsigned long flags;
859 struct kmemleak_object *object;
860
861 object = find_and_get_object(ptr, 0);
862 if (!object) {
863 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
864 return;
865 }
866
867 spin_lock_irqsave(&object->lock, flags);
868 object->flags |= OBJECT_NO_SCAN;
869 spin_unlock_irqrestore(&object->lock, flags);
870 put_object(object);
871}
872
873/**
874 * kmemleak_alloc - register a newly allocated object
875 * @ptr: pointer to beginning of the object
876 * @size: size of the object
877 * @min_count: minimum number of references to this object. If during memory
878 * scanning a number of references less than @min_count is found,
879 * the object is reported as a memory leak. If @min_count is 0,
880 * the object is never reported as a leak. If @min_count is -1,
881 * the object is ignored (not scanned and not reported as a leak)
882 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
883 *
884 * This function is called from the kernel allocators when a new object
885 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
886 */
887void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
888 gfp_t gfp)
889{
890 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
891
892 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
893 create_object((unsigned long)ptr, size, min_count, gfp);
894}
895EXPORT_SYMBOL_GPL(kmemleak_alloc);
896
897/**
898 * kmemleak_alloc_percpu - register a newly allocated __percpu object
899 * @ptr: __percpu pointer to beginning of the object
900 * @size: size of the object
901 * @gfp: flags used for kmemleak internal memory allocations
902 *
903 * This function is called from the kernel percpu allocator when a new object
904 * (memory block) is allocated (alloc_percpu).
905 */
906void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
907 gfp_t gfp)
908{
909 unsigned int cpu;
910
911 pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
912
913 /*
914 * Percpu allocations are only scanned and not reported as leaks
915 * (min_count is set to 0).
916 */
917 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
918 for_each_possible_cpu(cpu)
919 create_object((unsigned long)per_cpu_ptr(ptr, cpu),
920 size, 0, gfp);
921}
922EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
923
924/**
925 * kmemleak_vmalloc - register a newly vmalloc'ed object
926 * @area: pointer to vm_struct
927 * @size: size of the object
928 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
929 *
930 * This function is called from the vmalloc() kernel allocator when a new
931 * object (memory block) is allocated.
932 */
933void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
934{
935 pr_debug("%s(0x%p, %zu)\n", __func__, area, size);
936
937 /*
938 * A min_count = 2 is needed because vm_struct contains a reference to
939 * the virtual address of the vmalloc'ed block.
940 */
941 if (kmemleak_enabled) {
942 create_object((unsigned long)area->addr, size, 2, gfp);
943 object_set_excess_ref((unsigned long)area,
944 (unsigned long)area->addr);
945 }
946}
947EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
948
949/**
950 * kmemleak_free - unregister a previously registered object
951 * @ptr: pointer to beginning of the object
952 *
953 * This function is called from the kernel allocators when an object (memory
954 * block) is freed (kmem_cache_free, kfree, vfree etc.).
955 */
956void __ref kmemleak_free(const void *ptr)
957{
958 pr_debug("%s(0x%p)\n", __func__, ptr);
959
960 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
961 delete_object_full((unsigned long)ptr);
962}
963EXPORT_SYMBOL_GPL(kmemleak_free);
964
965/**
966 * kmemleak_free_part - partially unregister a previously registered object
967 * @ptr: pointer to the beginning or inside the object. This also
968 * represents the start of the range to be freed
969 * @size: size to be unregistered
970 *
971 * This function is called when only a part of a memory block is freed
972 * (usually from the bootmem allocator).
973 */
974void __ref kmemleak_free_part(const void *ptr, size_t size)
975{
976 pr_debug("%s(0x%p)\n", __func__, ptr);
977
978 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
979 delete_object_part((unsigned long)ptr, size);
980}
981EXPORT_SYMBOL_GPL(kmemleak_free_part);
982
983/**
984 * kmemleak_free_percpu - unregister a previously registered __percpu object
985 * @ptr: __percpu pointer to beginning of the object
986 *
987 * This function is called from the kernel percpu allocator when an object
988 * (memory block) is freed (free_percpu).
989 */
990void __ref kmemleak_free_percpu(const void __percpu *ptr)
991{
992 unsigned int cpu;
993
994 pr_debug("%s(0x%p)\n", __func__, ptr);
995
996 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
997 for_each_possible_cpu(cpu)
998 delete_object_full((unsigned long)per_cpu_ptr(ptr,
999 cpu));
1000}
1001EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1002
1003/**
1004 * kmemleak_update_trace - update object allocation stack trace
1005 * @ptr: pointer to beginning of the object
1006 *
1007 * Override the object allocation stack trace for cases where the actual
1008 * allocation place is not always useful.
1009 */
1010void __ref kmemleak_update_trace(const void *ptr)
1011{
1012 struct kmemleak_object *object;
1013 unsigned long flags;
1014
1015 pr_debug("%s(0x%p)\n", __func__, ptr);
1016
1017 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1018 return;
1019
1020 object = find_and_get_object((unsigned long)ptr, 1);
1021 if (!object) {
1022#ifdef DEBUG
1023 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1024 ptr);
1025#endif
1026 return;
1027 }
1028
1029 spin_lock_irqsave(&object->lock, flags);
1030 object->trace_len = __save_stack_trace(object->trace);
1031 spin_unlock_irqrestore(&object->lock, flags);
1032
1033 put_object(object);
1034}
1035EXPORT_SYMBOL(kmemleak_update_trace);
1036
1037/**
1038 * kmemleak_not_leak - mark an allocated object as false positive
1039 * @ptr: pointer to beginning of the object
1040 *
1041 * Calling this function on an object will cause the memory block to no longer
1042 * be reported as leak and always be scanned.
1043 */
1044void __ref kmemleak_not_leak(const void *ptr)
1045{
1046 pr_debug("%s(0x%p)\n", __func__, ptr);
1047
1048 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1049 make_gray_object((unsigned long)ptr);
1050}
1051EXPORT_SYMBOL(kmemleak_not_leak);
1052
1053/**
1054 * kmemleak_ignore - ignore an allocated object
1055 * @ptr: pointer to beginning of the object
1056 *
1057 * Calling this function on an object will cause the memory block to be
1058 * ignored (not scanned and not reported as a leak). This is usually done when
1059 * it is known that the corresponding block is not a leak and does not contain
1060 * any references to other allocated memory blocks.
1061 */
1062void __ref kmemleak_ignore(const void *ptr)
1063{
1064 pr_debug("%s(0x%p)\n", __func__, ptr);
1065
1066 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1067 make_black_object((unsigned long)ptr);
1068}
1069EXPORT_SYMBOL(kmemleak_ignore);
1070
1071/**
1072 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1073 * @ptr: pointer to beginning or inside the object. This also
1074 * represents the start of the scan area
1075 * @size: size of the scan area
1076 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1077 *
1078 * This function is used when it is known that only certain parts of an object
1079 * contain references to other objects. Kmemleak will only scan these areas
1080 * reducing the number false negatives.
1081 */
1082void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1083{
1084 pr_debug("%s(0x%p)\n", __func__, ptr);
1085
1086 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1087 add_scan_area((unsigned long)ptr, size, gfp);
1088}
1089EXPORT_SYMBOL(kmemleak_scan_area);
1090
1091/**
1092 * kmemleak_no_scan - do not scan an allocated object
1093 * @ptr: pointer to beginning of the object
1094 *
1095 * This function notifies kmemleak not to scan the given memory block. Useful
1096 * in situations where it is known that the given object does not contain any
1097 * references to other objects. Kmemleak will not scan such objects reducing
1098 * the number of false negatives.
1099 */
1100void __ref kmemleak_no_scan(const void *ptr)
1101{
1102 pr_debug("%s(0x%p)\n", __func__, ptr);
1103
1104 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1105 object_no_scan((unsigned long)ptr);
1106}
1107EXPORT_SYMBOL(kmemleak_no_scan);
1108
1109/**
1110 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1111 * address argument
1112 * @phys: physical address of the object
1113 * @size: size of the object
1114 * @min_count: minimum number of references to this object.
1115 * See kmemleak_alloc()
1116 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1117 */
1118void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, int min_count,
1119 gfp_t gfp)
1120{
1121 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1122 kmemleak_alloc(__va(phys), size, min_count, gfp);
1123}
1124EXPORT_SYMBOL(kmemleak_alloc_phys);
1125
1126/**
1127 * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1128 * physical address argument
1129 * @phys: physical address if the beginning or inside an object. This
1130 * also represents the start of the range to be freed
1131 * @size: size to be unregistered
1132 */
1133void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1134{
1135 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1136 kmemleak_free_part(__va(phys), size);
1137}
1138EXPORT_SYMBOL(kmemleak_free_part_phys);
1139
1140/**
1141 * kmemleak_not_leak_phys - similar to kmemleak_not_leak but taking a physical
1142 * address argument
1143 * @phys: physical address of the object
1144 */
1145void __ref kmemleak_not_leak_phys(phys_addr_t phys)
1146{
1147 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1148 kmemleak_not_leak(__va(phys));
1149}
1150EXPORT_SYMBOL(kmemleak_not_leak_phys);
1151
1152/**
1153 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1154 * address argument
1155 * @phys: physical address of the object
1156 */
1157void __ref kmemleak_ignore_phys(phys_addr_t phys)
1158{
1159 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1160 kmemleak_ignore(__va(phys));
1161}
1162EXPORT_SYMBOL(kmemleak_ignore_phys);
1163
1164/*
1165 * Update an object's checksum and return true if it was modified.
1166 */
1167static bool update_checksum(struct kmemleak_object *object)
1168{
1169 u32 old_csum = object->checksum;
1170
1171 kasan_disable_current();
1172 object->checksum = crc32(0, (void *)object->pointer, object->size);
1173 kasan_enable_current();
1174
1175 return object->checksum != old_csum;
1176}
1177
1178/*
1179 * Update an object's references. object->lock must be held by the caller.
1180 */
1181static void update_refs(struct kmemleak_object *object)
1182{
1183 if (!color_white(object)) {
1184 /* non-orphan, ignored or new */
1185 return;
1186 }
1187
1188 /*
1189 * Increase the object's reference count (number of pointers to the
1190 * memory block). If this count reaches the required minimum, the
1191 * object's color will become gray and it will be added to the
1192 * gray_list.
1193 */
1194 object->count++;
1195 if (color_gray(object)) {
1196 /* put_object() called when removing from gray_list */
1197 WARN_ON(!get_object(object));
1198 list_add_tail(&object->gray_list, &gray_list);
1199 }
1200}
1201
1202/*
1203 * Memory scanning is a long process and it needs to be interruptable. This
1204 * function checks whether such interrupt condition occurred.
1205 */
1206static int scan_should_stop(void)
1207{
1208 if (!kmemleak_enabled)
1209 return 1;
1210
1211 /*
1212 * This function may be called from either process or kthread context,
1213 * hence the need to check for both stop conditions.
1214 */
1215 if (current->mm)
1216 return signal_pending(current);
1217 else
1218 return kthread_should_stop();
1219
1220 return 0;
1221}
1222
1223/*
1224 * Scan a memory block (exclusive range) for valid pointers and add those
1225 * found to the gray list.
1226 */
1227static void scan_block(void *_start, void *_end,
1228 struct kmemleak_object *scanned)
1229{
1230 unsigned long *ptr;
1231 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1232 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1233 unsigned long flags;
1234 unsigned long untagged_ptr;
1235
1236 read_lock_irqsave(&kmemleak_lock, flags);
1237 for (ptr = start; ptr < end; ptr++) {
1238 struct kmemleak_object *object;
1239 unsigned long pointer;
1240 unsigned long excess_ref;
1241
1242 if (scan_should_stop())
1243 break;
1244
1245 kasan_disable_current();
1246 pointer = *ptr;
1247 kasan_enable_current();
1248
1249 untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
1250 if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
1251 continue;
1252
1253 /*
1254 * No need for get_object() here since we hold kmemleak_lock.
1255 * object->use_count cannot be dropped to 0 while the object
1256 * is still present in object_tree_root and object_list
1257 * (with updates protected by kmemleak_lock).
1258 */
1259 object = lookup_object(pointer, 1);
1260 if (!object)
1261 continue;
1262 if (object == scanned)
1263 /* self referenced, ignore */
1264 continue;
1265
1266 /*
1267 * Avoid the lockdep recursive warning on object->lock being
1268 * previously acquired in scan_object(). These locks are
1269 * enclosed by scan_mutex.
1270 */
1271 spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1272 /* only pass surplus references (object already gray) */
1273 if (color_gray(object)) {
1274 excess_ref = object->excess_ref;
1275 /* no need for update_refs() if object already gray */
1276 } else {
1277 excess_ref = 0;
1278 update_refs(object);
1279 }
1280 spin_unlock(&object->lock);
1281
1282 if (excess_ref) {
1283 object = lookup_object(excess_ref, 0);
1284 if (!object)
1285 continue;
1286 if (object == scanned)
1287 /* circular reference, ignore */
1288 continue;
1289 spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1290 update_refs(object);
1291 spin_unlock(&object->lock);
1292 }
1293 }
1294 read_unlock_irqrestore(&kmemleak_lock, flags);
1295}
1296
1297/*
1298 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1299 */
1300#ifdef CONFIG_SMP
1301static void scan_large_block(void *start, void *end)
1302{
1303 void *next;
1304
1305 while (start < end) {
1306 next = min(start + MAX_SCAN_SIZE, end);
1307 scan_block(start, next, NULL);
1308 start = next;
1309 cond_resched();
1310 }
1311}
1312#endif
1313
1314/*
1315 * Scan a memory block corresponding to a kmemleak_object. A condition is
1316 * that object->use_count >= 1.
1317 */
1318static void scan_object(struct kmemleak_object *object)
1319{
1320 struct kmemleak_scan_area *area;
1321 unsigned long flags;
1322
1323 /*
1324 * Once the object->lock is acquired, the corresponding memory block
1325 * cannot be freed (the same lock is acquired in delete_object).
1326 */
1327 spin_lock_irqsave(&object->lock, flags);
1328 if (object->flags & OBJECT_NO_SCAN)
1329 goto out;
1330 if (!(object->flags & OBJECT_ALLOCATED))
1331 /* already freed object */
1332 goto out;
1333 if (hlist_empty(&object->area_list) ||
1334 object->flags & OBJECT_FULL_SCAN) {
1335 void *start = (void *)object->pointer;
1336 void *end = (void *)(object->pointer + object->size);
1337 void *next;
1338
1339 do {
1340 next = min(start + MAX_SCAN_SIZE, end);
1341 scan_block(start, next, object);
1342
1343 start = next;
1344 if (start >= end)
1345 break;
1346
1347 spin_unlock_irqrestore(&object->lock, flags);
1348 cond_resched();
1349 spin_lock_irqsave(&object->lock, flags);
1350 } while (object->flags & OBJECT_ALLOCATED);
1351 } else
1352 hlist_for_each_entry(area, &object->area_list, node)
1353 scan_block((void *)area->start,
1354 (void *)(area->start + area->size),
1355 object);
1356out:
1357 spin_unlock_irqrestore(&object->lock, flags);
1358}
1359
1360/*
1361 * Scan the objects already referenced (gray objects). More objects will be
1362 * referenced and, if there are no memory leaks, all the objects are scanned.
1363 */
1364static void scan_gray_list(void)
1365{
1366 struct kmemleak_object *object, *tmp;
1367
1368 /*
1369 * The list traversal is safe for both tail additions and removals
1370 * from inside the loop. The kmemleak objects cannot be freed from
1371 * outside the loop because their use_count was incremented.
1372 */
1373 object = list_entry(gray_list.next, typeof(*object), gray_list);
1374 while (&object->gray_list != &gray_list) {
1375 cond_resched();
1376
1377 /* may add new objects to the list */
1378 if (!scan_should_stop())
1379 scan_object(object);
1380
1381 tmp = list_entry(object->gray_list.next, typeof(*object),
1382 gray_list);
1383
1384 /* remove the object from the list and release it */
1385 list_del(&object->gray_list);
1386 put_object(object);
1387
1388 object = tmp;
1389 }
1390 WARN_ON(!list_empty(&gray_list));
1391}
1392
1393/*
1394 * Scan data sections and all the referenced memory blocks allocated via the
1395 * kernel's standard allocators. This function must be called with the
1396 * scan_mutex held.
1397 */
1398static void kmemleak_scan(void)
1399{
1400 unsigned long flags;
1401 struct kmemleak_object *object;
1402 int i;
1403 int new_leaks = 0;
1404
1405 jiffies_last_scan = jiffies;
1406
1407 /* prepare the kmemleak_object's */
1408 rcu_read_lock();
1409 list_for_each_entry_rcu(object, &object_list, object_list) {
1410 spin_lock_irqsave(&object->lock, flags);
1411#ifdef DEBUG
1412 /*
1413 * With a few exceptions there should be a maximum of
1414 * 1 reference to any object at this point.
1415 */
1416 if (atomic_read(&object->use_count) > 1) {
1417 pr_debug("object->use_count = %d\n",
1418 atomic_read(&object->use_count));
1419 dump_object_info(object);
1420 }
1421#endif
1422 /* reset the reference count (whiten the object) */
1423 object->count = 0;
1424 if (color_gray(object) && get_object(object))
1425 list_add_tail(&object->gray_list, &gray_list);
1426
1427 spin_unlock_irqrestore(&object->lock, flags);
1428 }
1429 rcu_read_unlock();
1430
1431#ifdef CONFIG_SMP
1432 /* per-cpu sections scanning */
1433 for_each_possible_cpu(i)
1434 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1435 __per_cpu_end + per_cpu_offset(i));
1436#endif
1437
1438 /*
1439 * Struct page scanning for each node.
1440 */
1441 get_online_mems();
1442 for_each_online_node(i) {
1443 unsigned long start_pfn = node_start_pfn(i);
1444 unsigned long end_pfn = node_end_pfn(i);
1445 unsigned long pfn;
1446
1447 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1448 struct page *page = pfn_to_online_page(pfn);
1449
1450 if (!page)
1451 continue;
1452
1453 /* only scan pages belonging to this node */
1454 if (page_to_nid(page) != i)
1455 continue;
1456 /* only scan if page is in use */
1457 if (page_count(page) == 0)
1458 continue;
1459 scan_block(page, page + 1, NULL);
1460 if (!(pfn & 63))
1461 cond_resched();
1462 }
1463 }
1464 put_online_mems();
1465
1466 /*
1467 * Scanning the task stacks (may introduce false negatives).
1468 */
1469 if (kmemleak_stack_scan) {
1470 struct task_struct *p, *g;
1471
1472 read_lock(&tasklist_lock);
1473 do_each_thread(g, p) {
1474 void *stack = try_get_task_stack(p);
1475 if (stack) {
1476 scan_block(stack, stack + THREAD_SIZE, NULL);
1477 put_task_stack(p);
1478 }
1479 } while_each_thread(g, p);
1480 read_unlock(&tasklist_lock);
1481 }
1482
1483 /*
1484 * Scan the objects already referenced from the sections scanned
1485 * above.
1486 */
1487 scan_gray_list();
1488
1489 /*
1490 * Check for new or unreferenced objects modified since the previous
1491 * scan and color them gray until the next scan.
1492 */
1493 rcu_read_lock();
1494 list_for_each_entry_rcu(object, &object_list, object_list) {
1495 spin_lock_irqsave(&object->lock, flags);
1496 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1497 && update_checksum(object) && get_object(object)) {
1498 /* color it gray temporarily */
1499 object->count = object->min_count;
1500 list_add_tail(&object->gray_list, &gray_list);
1501 }
1502 spin_unlock_irqrestore(&object->lock, flags);
1503 }
1504 rcu_read_unlock();
1505
1506 /*
1507 * Re-scan the gray list for modified unreferenced objects.
1508 */
1509 scan_gray_list();
1510
1511 /*
1512 * If scanning was stopped do not report any new unreferenced objects.
1513 */
1514 if (scan_should_stop())
1515 return;
1516
1517 /*
1518 * Scanning result reporting.
1519 */
1520 rcu_read_lock();
1521 list_for_each_entry_rcu(object, &object_list, object_list) {
1522 spin_lock_irqsave(&object->lock, flags);
1523 if (unreferenced_object(object) &&
1524 !(object->flags & OBJECT_REPORTED)) {
1525 object->flags |= OBJECT_REPORTED;
1526
1527 if (kmemleak_verbose)
1528 print_unreferenced(NULL, object);
1529
1530 new_leaks++;
1531 }
1532 spin_unlock_irqrestore(&object->lock, flags);
1533 }
1534 rcu_read_unlock();
1535
1536 if (new_leaks) {
1537 kmemleak_found_leaks = true;
1538
1539 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1540 new_leaks);
1541 }
1542
1543}
1544
1545/*
1546 * Thread function performing automatic memory scanning. Unreferenced objects
1547 * at the end of a memory scan are reported but only the first time.
1548 */
1549static int kmemleak_scan_thread(void *arg)
1550{
1551 static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
1552
1553 pr_info("Automatic memory scanning thread started\n");
1554 set_user_nice(current, 10);
1555
1556 /*
1557 * Wait before the first scan to allow the system to fully initialize.
1558 */
1559 if (first_run) {
1560 signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
1561 first_run = 0;
1562 while (timeout && !kthread_should_stop())
1563 timeout = schedule_timeout_interruptible(timeout);
1564 }
1565
1566 while (!kthread_should_stop()) {
1567 signed long timeout = jiffies_scan_wait;
1568
1569 mutex_lock(&scan_mutex);
1570 kmemleak_scan();
1571 mutex_unlock(&scan_mutex);
1572
1573 /* wait before the next scan */
1574 while (timeout && !kthread_should_stop())
1575 timeout = schedule_timeout_interruptible(timeout);
1576 }
1577
1578 pr_info("Automatic memory scanning thread ended\n");
1579
1580 return 0;
1581}
1582
1583/*
1584 * Start the automatic memory scanning thread. This function must be called
1585 * with the scan_mutex held.
1586 */
1587static void start_scan_thread(void)
1588{
1589 if (scan_thread)
1590 return;
1591 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1592 if (IS_ERR(scan_thread)) {
1593 pr_warn("Failed to create the scan thread\n");
1594 scan_thread = NULL;
1595 }
1596}
1597
1598/*
1599 * Stop the automatic memory scanning thread.
1600 */
1601static void stop_scan_thread(void)
1602{
1603 if (scan_thread) {
1604 kthread_stop(scan_thread);
1605 scan_thread = NULL;
1606 }
1607}
1608
1609/*
1610 * Iterate over the object_list and return the first valid object at or after
1611 * the required position with its use_count incremented. The function triggers
1612 * a memory scanning when the pos argument points to the first position.
1613 */
1614static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1615{
1616 struct kmemleak_object *object;
1617 loff_t n = *pos;
1618 int err;
1619
1620 err = mutex_lock_interruptible(&scan_mutex);
1621 if (err < 0)
1622 return ERR_PTR(err);
1623
1624 rcu_read_lock();
1625 list_for_each_entry_rcu(object, &object_list, object_list) {
1626 if (n-- > 0)
1627 continue;
1628 if (get_object(object))
1629 goto out;
1630 }
1631 object = NULL;
1632out:
1633 return object;
1634}
1635
1636/*
1637 * Return the next object in the object_list. The function decrements the
1638 * use_count of the previous object and increases that of the next one.
1639 */
1640static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1641{
1642 struct kmemleak_object *prev_obj = v;
1643 struct kmemleak_object *next_obj = NULL;
1644 struct kmemleak_object *obj = prev_obj;
1645
1646 ++(*pos);
1647
1648 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1649 if (get_object(obj)) {
1650 next_obj = obj;
1651 break;
1652 }
1653 }
1654
1655 put_object(prev_obj);
1656 return next_obj;
1657}
1658
1659/*
1660 * Decrement the use_count of the last object required, if any.
1661 */
1662static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1663{
1664 if (!IS_ERR(v)) {
1665 /*
1666 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1667 * waiting was interrupted, so only release it if !IS_ERR.
1668 */
1669 rcu_read_unlock();
1670 mutex_unlock(&scan_mutex);
1671 if (v)
1672 put_object(v);
1673 }
1674}
1675
1676/*
1677 * Print the information for an unreferenced object to the seq file.
1678 */
1679static int kmemleak_seq_show(struct seq_file *seq, void *v)
1680{
1681 struct kmemleak_object *object = v;
1682 unsigned long flags;
1683
1684 spin_lock_irqsave(&object->lock, flags);
1685 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1686 print_unreferenced(seq, object);
1687 spin_unlock_irqrestore(&object->lock, flags);
1688 return 0;
1689}
1690
1691static const struct seq_operations kmemleak_seq_ops = {
1692 .start = kmemleak_seq_start,
1693 .next = kmemleak_seq_next,
1694 .stop = kmemleak_seq_stop,
1695 .show = kmemleak_seq_show,
1696};
1697
1698static int kmemleak_open(struct inode *inode, struct file *file)
1699{
1700 return seq_open(file, &kmemleak_seq_ops);
1701}
1702
1703static int dump_str_object_info(const char *str)
1704{
1705 unsigned long flags;
1706 struct kmemleak_object *object;
1707 unsigned long addr;
1708
1709 if (kstrtoul(str, 0, &addr))
1710 return -EINVAL;
1711 object = find_and_get_object(addr, 0);
1712 if (!object) {
1713 pr_info("Unknown object at 0x%08lx\n", addr);
1714 return -EINVAL;
1715 }
1716
1717 spin_lock_irqsave(&object->lock, flags);
1718 dump_object_info(object);
1719 spin_unlock_irqrestore(&object->lock, flags);
1720
1721 put_object(object);
1722 return 0;
1723}
1724
1725/*
1726 * We use grey instead of black to ensure we can do future scans on the same
1727 * objects. If we did not do future scans these black objects could
1728 * potentially contain references to newly allocated objects in the future and
1729 * we'd end up with false positives.
1730 */
1731static void kmemleak_clear(void)
1732{
1733 struct kmemleak_object *object;
1734 unsigned long flags;
1735
1736 rcu_read_lock();
1737 list_for_each_entry_rcu(object, &object_list, object_list) {
1738 spin_lock_irqsave(&object->lock, flags);
1739 if ((object->flags & OBJECT_REPORTED) &&
1740 unreferenced_object(object))
1741 __paint_it(object, KMEMLEAK_GREY);
1742 spin_unlock_irqrestore(&object->lock, flags);
1743 }
1744 rcu_read_unlock();
1745
1746 kmemleak_found_leaks = false;
1747}
1748
1749static void __kmemleak_do_cleanup(void);
1750
1751/*
1752 * File write operation to configure kmemleak at run-time. The following
1753 * commands can be written to the /sys/kernel/debug/kmemleak file:
1754 * off - disable kmemleak (irreversible)
1755 * stack=on - enable the task stacks scanning
1756 * stack=off - disable the tasks stacks scanning
1757 * scan=on - start the automatic memory scanning thread
1758 * scan=off - stop the automatic memory scanning thread
1759 * scan=... - set the automatic memory scanning period in seconds (0 to
1760 * disable it)
1761 * scan - trigger a memory scan
1762 * clear - mark all current reported unreferenced kmemleak objects as
1763 * grey to ignore printing them, or free all kmemleak objects
1764 * if kmemleak has been disabled.
1765 * dump=... - dump information about the object found at the given address
1766 */
1767static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1768 size_t size, loff_t *ppos)
1769{
1770 char buf[64];
1771 int buf_size;
1772 int ret;
1773
1774 buf_size = min(size, (sizeof(buf) - 1));
1775 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1776 return -EFAULT;
1777 buf[buf_size] = 0;
1778
1779 ret = mutex_lock_interruptible(&scan_mutex);
1780 if (ret < 0)
1781 return ret;
1782
1783 if (strncmp(buf, "clear", 5) == 0) {
1784 if (kmemleak_enabled)
1785 kmemleak_clear();
1786 else
1787 __kmemleak_do_cleanup();
1788 goto out;
1789 }
1790
1791 if (!kmemleak_enabled) {
1792 ret = -EPERM;
1793 goto out;
1794 }
1795
1796 if (strncmp(buf, "off", 3) == 0)
1797 kmemleak_disable();
1798 else if (strncmp(buf, "stack=on", 8) == 0)
1799 kmemleak_stack_scan = 1;
1800 else if (strncmp(buf, "stack=off", 9) == 0)
1801 kmemleak_stack_scan = 0;
1802 else if (strncmp(buf, "scan=on", 7) == 0)
1803 start_scan_thread();
1804 else if (strncmp(buf, "scan=off", 8) == 0)
1805 stop_scan_thread();
1806 else if (strncmp(buf, "scan=", 5) == 0) {
1807 unsigned long secs;
1808
1809 ret = kstrtoul(buf + 5, 0, &secs);
1810 if (ret < 0)
1811 goto out;
1812 stop_scan_thread();
1813 if (secs) {
1814 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1815 start_scan_thread();
1816 }
1817 } else if (strncmp(buf, "scan", 4) == 0)
1818 kmemleak_scan();
1819 else if (strncmp(buf, "dump=", 5) == 0)
1820 ret = dump_str_object_info(buf + 5);
1821 else
1822 ret = -EINVAL;
1823
1824out:
1825 mutex_unlock(&scan_mutex);
1826 if (ret < 0)
1827 return ret;
1828
1829 /* ignore the rest of the buffer, only one command at a time */
1830 *ppos += size;
1831 return size;
1832}
1833
1834static const struct file_operations kmemleak_fops = {
1835 .owner = THIS_MODULE,
1836 .open = kmemleak_open,
1837 .read = seq_read,
1838 .write = kmemleak_write,
1839 .llseek = seq_lseek,
1840 .release = seq_release,
1841};
1842
1843static void __kmemleak_do_cleanup(void)
1844{
1845 struct kmemleak_object *object, *tmp;
1846
1847 /*
1848 * Kmemleak has already been disabled, no need for RCU list traversal
1849 * or kmemleak_lock held.
1850 */
1851 list_for_each_entry_safe(object, tmp, &object_list, object_list) {
1852 __remove_object(object);
1853 __delete_object(object);
1854 }
1855}
1856
1857/*
1858 * Stop the memory scanning thread and free the kmemleak internal objects if
1859 * no previous scan thread (otherwise, kmemleak may still have some useful
1860 * information on memory leaks).
1861 */
1862static void kmemleak_do_cleanup(struct work_struct *work)
1863{
1864 stop_scan_thread();
1865
1866 mutex_lock(&scan_mutex);
1867 /*
1868 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
1869 * longer track object freeing. Ordering of the scan thread stopping and
1870 * the memory accesses below is guaranteed by the kthread_stop()
1871 * function.
1872 */
1873 kmemleak_free_enabled = 0;
1874 mutex_unlock(&scan_mutex);
1875
1876 if (!kmemleak_found_leaks)
1877 __kmemleak_do_cleanup();
1878 else
1879 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
1880}
1881
1882static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1883
1884/*
1885 * Disable kmemleak. No memory allocation/freeing will be traced once this
1886 * function is called. Disabling kmemleak is an irreversible operation.
1887 */
1888static void kmemleak_disable(void)
1889{
1890 /* atomically check whether it was already invoked */
1891 if (cmpxchg(&kmemleak_error, 0, 1))
1892 return;
1893
1894 /* stop any memory operation tracing */
1895 kmemleak_enabled = 0;
1896
1897 /* check whether it is too early for a kernel thread */
1898 if (kmemleak_initialized)
1899 schedule_work(&cleanup_work);
1900 else
1901 kmemleak_free_enabled = 0;
1902
1903 pr_info("Kernel memory leak detector disabled\n");
1904}
1905
1906/*
1907 * Allow boot-time kmemleak disabling (enabled by default).
1908 */
1909static int __init kmemleak_boot_config(char *str)
1910{
1911 if (!str)
1912 return -EINVAL;
1913 if (strcmp(str, "off") == 0)
1914 kmemleak_disable();
1915 else if (strcmp(str, "on") == 0)
1916 kmemleak_skip_disable = 1;
1917 else
1918 return -EINVAL;
1919 return 0;
1920}
1921early_param("kmemleak", kmemleak_boot_config);
1922
1923/*
1924 * Kmemleak initialization.
1925 */
1926void __init kmemleak_init(void)
1927{
1928#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
1929 if (!kmemleak_skip_disable) {
1930 kmemleak_disable();
1931 return;
1932 }
1933#endif
1934
1935 if (kmemleak_error)
1936 return;
1937
1938 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1939 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1940
1941 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1942 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1943
1944 /* register the data/bss sections */
1945 create_object((unsigned long)_sdata, _edata - _sdata,
1946 KMEMLEAK_GREY, GFP_ATOMIC);
1947 create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
1948 KMEMLEAK_GREY, GFP_ATOMIC);
1949 /* only register .data..ro_after_init if not within .data */
1950 if (__start_ro_after_init < _sdata || __end_ro_after_init > _edata)
1951 create_object((unsigned long)__start_ro_after_init,
1952 __end_ro_after_init - __start_ro_after_init,
1953 KMEMLEAK_GREY, GFP_ATOMIC);
1954}
1955
1956/*
1957 * Late initialization function.
1958 */
1959static int __init kmemleak_late_init(void)
1960{
1961 kmemleak_initialized = 1;
1962
1963 debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops);
1964
1965 if (kmemleak_error) {
1966 /*
1967 * Some error occurred and kmemleak was disabled. There is a
1968 * small chance that kmemleak_disable() was called immediately
1969 * after setting kmemleak_initialized and we may end up with
1970 * two clean-up threads but serialized by scan_mutex.
1971 */
1972 schedule_work(&cleanup_work);
1973 return -ENOMEM;
1974 }
1975
1976 if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
1977 mutex_lock(&scan_mutex);
1978 start_scan_thread();
1979 mutex_unlock(&scan_mutex);
1980 }
1981
1982 pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
1983 mem_pool_free_count);
1984
1985 return 0;
1986}
1987late_initcall(kmemleak_late_init);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * mm/kmemleak.c
4 *
5 * Copyright (C) 2008 ARM Limited
6 * Written by Catalin Marinas <catalin.marinas@arm.com>
7 *
8 * For more information on the algorithm and kmemleak usage, please see
9 * Documentation/dev-tools/kmemleak.rst.
10 *
11 * Notes on locking
12 * ----------------
13 *
14 * The following locks and mutexes are used by kmemleak:
15 *
16 * - kmemleak_lock (raw_spinlock_t): protects the object_list as well as
17 * del_state modifications and accesses to the object trees
18 * (object_tree_root, object_phys_tree_root, object_percpu_tree_root). The
19 * object_list is the main list holding the metadata (struct
20 * kmemleak_object) for the allocated memory blocks. The object trees are
21 * red black trees used to look-up metadata based on a pointer to the
22 * corresponding memory block. The kmemleak_object structures are added to
23 * the object_list and the object tree root in the create_object() function
24 * called from the kmemleak_alloc{,_phys,_percpu}() callback and removed in
25 * delete_object() called from the kmemleak_free{,_phys,_percpu}() callback
26 * - kmemleak_object.lock (raw_spinlock_t): protects a kmemleak_object.
27 * Accesses to the metadata (e.g. count) are protected by this lock. Note
28 * that some members of this structure may be protected by other means
29 * (atomic or kmemleak_lock). This lock is also held when scanning the
30 * corresponding memory block to avoid the kernel freeing it via the
31 * kmemleak_free() callback. This is less heavyweight than holding a global
32 * lock like kmemleak_lock during scanning.
33 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
34 * unreferenced objects at a time. The gray_list contains the objects which
35 * are already referenced or marked as false positives and need to be
36 * scanned. This list is only modified during a scanning episode when the
37 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
38 * Note that the kmemleak_object.use_count is incremented when an object is
39 * added to the gray_list and therefore cannot be freed. This mutex also
40 * prevents multiple users of the "kmemleak" debugfs file together with
41 * modifications to the memory scanning parameters including the scan_thread
42 * pointer
43 *
44 * Locks and mutexes are acquired/nested in the following order:
45 *
46 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
47 *
48 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
49 * regions.
50 *
51 * The kmemleak_object structures have a use_count incremented or decremented
52 * using the get_object()/put_object() functions. When the use_count becomes
53 * 0, this count can no longer be incremented and put_object() schedules the
54 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
55 * function must be protected by rcu_read_lock() to avoid accessing a freed
56 * structure.
57 */
58
59#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
60
61#include <linux/init.h>
62#include <linux/kernel.h>
63#include <linux/list.h>
64#include <linux/sched/signal.h>
65#include <linux/sched/task.h>
66#include <linux/sched/task_stack.h>
67#include <linux/jiffies.h>
68#include <linux/delay.h>
69#include <linux/export.h>
70#include <linux/kthread.h>
71#include <linux/rbtree.h>
72#include <linux/fs.h>
73#include <linux/debugfs.h>
74#include <linux/seq_file.h>
75#include <linux/cpumask.h>
76#include <linux/spinlock.h>
77#include <linux/module.h>
78#include <linux/mutex.h>
79#include <linux/rcupdate.h>
80#include <linux/stacktrace.h>
81#include <linux/stackdepot.h>
82#include <linux/cache.h>
83#include <linux/percpu.h>
84#include <linux/memblock.h>
85#include <linux/pfn.h>
86#include <linux/mmzone.h>
87#include <linux/slab.h>
88#include <linux/thread_info.h>
89#include <linux/err.h>
90#include <linux/uaccess.h>
91#include <linux/string.h>
92#include <linux/nodemask.h>
93#include <linux/mm.h>
94#include <linux/workqueue.h>
95#include <linux/crc32.h>
96
97#include <asm/sections.h>
98#include <asm/processor.h>
99#include <linux/atomic.h>
100
101#include <linux/kasan.h>
102#include <linux/kfence.h>
103#include <linux/kmemleak.h>
104#include <linux/memory_hotplug.h>
105
106/*
107 * Kmemleak configuration and common defines.
108 */
109#define MAX_TRACE 16 /* stack trace length */
110#define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
111#define SECS_FIRST_SCAN 60 /* delay before the first scan */
112#define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
113#define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
114
115#define BYTES_PER_POINTER sizeof(void *)
116
117/* GFP bitmask for kmemleak internal allocations */
118#define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC | \
119 __GFP_NOLOCKDEP)) | \
120 __GFP_NORETRY | __GFP_NOMEMALLOC | \
121 __GFP_NOWARN)
122
123/* scanning area inside a memory block */
124struct kmemleak_scan_area {
125 struct hlist_node node;
126 unsigned long start;
127 size_t size;
128};
129
130#define KMEMLEAK_GREY 0
131#define KMEMLEAK_BLACK -1
132
133/*
134 * Structure holding the metadata for each allocated memory block.
135 * Modifications to such objects should be made while holding the
136 * object->lock. Insertions or deletions from object_list, gray_list or
137 * rb_node are already protected by the corresponding locks or mutex (see
138 * the notes on locking above). These objects are reference-counted
139 * (use_count) and freed using the RCU mechanism.
140 */
141struct kmemleak_object {
142 raw_spinlock_t lock;
143 unsigned int flags; /* object status flags */
144 struct list_head object_list;
145 struct list_head gray_list;
146 struct rb_node rb_node;
147 struct rcu_head rcu; /* object_list lockless traversal */
148 /* object usage count; object freed when use_count == 0 */
149 atomic_t use_count;
150 unsigned int del_state; /* deletion state */
151 unsigned long pointer;
152 size_t size;
153 /* pass surplus references to this pointer */
154 unsigned long excess_ref;
155 /* minimum number of a pointers found before it is considered leak */
156 int min_count;
157 /* the total number of pointers found pointing to this object */
158 int count;
159 /* checksum for detecting modified objects */
160 u32 checksum;
161 /* memory ranges to be scanned inside an object (empty for all) */
162 struct hlist_head area_list;
163 depot_stack_handle_t trace_handle;
164 unsigned long jiffies; /* creation timestamp */
165 pid_t pid; /* pid of the current task */
166 char comm[TASK_COMM_LEN]; /* executable name */
167};
168
169/* flag representing the memory block allocation status */
170#define OBJECT_ALLOCATED (1 << 0)
171/* flag set after the first reporting of an unreference object */
172#define OBJECT_REPORTED (1 << 1)
173/* flag set to not scan the object */
174#define OBJECT_NO_SCAN (1 << 2)
175/* flag set to fully scan the object when scan_area allocation failed */
176#define OBJECT_FULL_SCAN (1 << 3)
177/* flag set for object allocated with physical address */
178#define OBJECT_PHYS (1 << 4)
179/* flag set for per-CPU pointers */
180#define OBJECT_PERCPU (1 << 5)
181
182/* set when __remove_object() called */
183#define DELSTATE_REMOVED (1 << 0)
184/* set to temporarily prevent deletion from object_list */
185#define DELSTATE_NO_DELETE (1 << 1)
186
187#define HEX_PREFIX " "
188/* number of bytes to print per line; must be 16 or 32 */
189#define HEX_ROW_SIZE 16
190/* number of bytes to print at a time (1, 2, 4, 8) */
191#define HEX_GROUP_SIZE 1
192/* include ASCII after the hex output */
193#define HEX_ASCII 1
194/* max number of lines to be printed */
195#define HEX_MAX_LINES 2
196
197/* the list of all allocated objects */
198static LIST_HEAD(object_list);
199/* the list of gray-colored objects (see color_gray comment below) */
200static LIST_HEAD(gray_list);
201/* memory pool allocation */
202static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE];
203static int mem_pool_free_count = ARRAY_SIZE(mem_pool);
204static LIST_HEAD(mem_pool_free_list);
205/* search tree for object boundaries */
206static struct rb_root object_tree_root = RB_ROOT;
207/* search tree for object (with OBJECT_PHYS flag) boundaries */
208static struct rb_root object_phys_tree_root = RB_ROOT;
209/* search tree for object (with OBJECT_PERCPU flag) boundaries */
210static struct rb_root object_percpu_tree_root = RB_ROOT;
211/* protecting the access to object_list, object_tree_root (or object_phys_tree_root) */
212static DEFINE_RAW_SPINLOCK(kmemleak_lock);
213
214/* allocation caches for kmemleak internal data */
215static struct kmem_cache *object_cache;
216static struct kmem_cache *scan_area_cache;
217
218/* set if tracing memory operations is enabled */
219static int kmemleak_enabled = 1;
220/* same as above but only for the kmemleak_free() callback */
221static int kmemleak_free_enabled = 1;
222/* set in the late_initcall if there were no errors */
223static int kmemleak_late_initialized;
224/* set if a kmemleak warning was issued */
225static int kmemleak_warning;
226/* set if a fatal kmemleak error has occurred */
227static int kmemleak_error;
228
229/* minimum and maximum address that may be valid pointers */
230static unsigned long min_addr = ULONG_MAX;
231static unsigned long max_addr;
232
233static struct task_struct *scan_thread;
234/* used to avoid reporting of recently allocated objects */
235static unsigned long jiffies_min_age;
236static unsigned long jiffies_last_scan;
237/* delay between automatic memory scannings */
238static unsigned long jiffies_scan_wait;
239/* enables or disables the task stacks scanning */
240static int kmemleak_stack_scan = 1;
241/* protects the memory scanning, parameters and debug/kmemleak file access */
242static DEFINE_MUTEX(scan_mutex);
243/* setting kmemleak=on, will set this var, skipping the disable */
244static int kmemleak_skip_disable;
245/* If there are leaks that can be reported */
246static bool kmemleak_found_leaks;
247
248static bool kmemleak_verbose;
249module_param_named(verbose, kmemleak_verbose, bool, 0600);
250
251static void kmemleak_disable(void);
252
253/*
254 * Print a warning and dump the stack trace.
255 */
256#define kmemleak_warn(x...) do { \
257 pr_warn(x); \
258 dump_stack(); \
259 kmemleak_warning = 1; \
260} while (0)
261
262/*
263 * Macro invoked when a serious kmemleak condition occurred and cannot be
264 * recovered from. Kmemleak will be disabled and further allocation/freeing
265 * tracing no longer available.
266 */
267#define kmemleak_stop(x...) do { \
268 kmemleak_warn(x); \
269 kmemleak_disable(); \
270} while (0)
271
272#define warn_or_seq_printf(seq, fmt, ...) do { \
273 if (seq) \
274 seq_printf(seq, fmt, ##__VA_ARGS__); \
275 else \
276 pr_warn(fmt, ##__VA_ARGS__); \
277} while (0)
278
279static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
280 int rowsize, int groupsize, const void *buf,
281 size_t len, bool ascii)
282{
283 if (seq)
284 seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
285 buf, len, ascii);
286 else
287 print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
288 rowsize, groupsize, buf, len, ascii);
289}
290
291/*
292 * Printing of the objects hex dump to the seq file. The number of lines to be
293 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
294 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
295 * with the object->lock held.
296 */
297static void hex_dump_object(struct seq_file *seq,
298 struct kmemleak_object *object)
299{
300 const u8 *ptr = (const u8 *)object->pointer;
301 size_t len;
302
303 if (WARN_ON_ONCE(object->flags & (OBJECT_PHYS | OBJECT_PERCPU)))
304 return;
305
306 /* limit the number of lines to HEX_MAX_LINES */
307 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
308
309 warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len);
310 kasan_disable_current();
311 warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
312 HEX_GROUP_SIZE, kasan_reset_tag((void *)ptr), len, HEX_ASCII);
313 kasan_enable_current();
314}
315
316/*
317 * Object colors, encoded with count and min_count:
318 * - white - orphan object, not enough references to it (count < min_count)
319 * - gray - not orphan, not marked as false positive (min_count == 0) or
320 * sufficient references to it (count >= min_count)
321 * - black - ignore, it doesn't contain references (e.g. text section)
322 * (min_count == -1). No function defined for this color.
323 * Newly created objects don't have any color assigned (object->count == -1)
324 * before the next memory scan when they become white.
325 */
326static bool color_white(const struct kmemleak_object *object)
327{
328 return object->count != KMEMLEAK_BLACK &&
329 object->count < object->min_count;
330}
331
332static bool color_gray(const struct kmemleak_object *object)
333{
334 return object->min_count != KMEMLEAK_BLACK &&
335 object->count >= object->min_count;
336}
337
338/*
339 * Objects are considered unreferenced only if their color is white, they have
340 * not be deleted and have a minimum age to avoid false positives caused by
341 * pointers temporarily stored in CPU registers.
342 */
343static bool unreferenced_object(struct kmemleak_object *object)
344{
345 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
346 time_before_eq(object->jiffies + jiffies_min_age,
347 jiffies_last_scan);
348}
349
350/*
351 * Printing of the unreferenced objects information to the seq file. The
352 * print_unreferenced function must be called with the object->lock held.
353 */
354static void print_unreferenced(struct seq_file *seq,
355 struct kmemleak_object *object)
356{
357 int i;
358 unsigned long *entries;
359 unsigned int nr_entries;
360
361 nr_entries = stack_depot_fetch(object->trace_handle, &entries);
362 warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
363 object->pointer, object->size);
364 warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n",
365 object->comm, object->pid, object->jiffies);
366 hex_dump_object(seq, object);
367 warn_or_seq_printf(seq, " backtrace (crc %x):\n", object->checksum);
368
369 for (i = 0; i < nr_entries; i++) {
370 void *ptr = (void *)entries[i];
371 warn_or_seq_printf(seq, " [<%pK>] %pS\n", ptr, ptr);
372 }
373}
374
375/*
376 * Print the kmemleak_object information. This function is used mainly for
377 * debugging special cases when kmemleak operations. It must be called with
378 * the object->lock held.
379 */
380static void dump_object_info(struct kmemleak_object *object)
381{
382 pr_notice("Object 0x%08lx (size %zu):\n",
383 object->pointer, object->size);
384 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
385 object->comm, object->pid, object->jiffies);
386 pr_notice(" min_count = %d\n", object->min_count);
387 pr_notice(" count = %d\n", object->count);
388 pr_notice(" flags = 0x%x\n", object->flags);
389 pr_notice(" checksum = %u\n", object->checksum);
390 pr_notice(" backtrace:\n");
391 if (object->trace_handle)
392 stack_depot_print(object->trace_handle);
393}
394
395static struct rb_root *object_tree(unsigned long objflags)
396{
397 if (objflags & OBJECT_PHYS)
398 return &object_phys_tree_root;
399 if (objflags & OBJECT_PERCPU)
400 return &object_percpu_tree_root;
401 return &object_tree_root;
402}
403
404/*
405 * Look-up a memory block metadata (kmemleak_object) in the object search
406 * tree based on a pointer value. If alias is 0, only values pointing to the
407 * beginning of the memory block are allowed. The kmemleak_lock must be held
408 * when calling this function.
409 */
410static struct kmemleak_object *__lookup_object(unsigned long ptr, int alias,
411 unsigned int objflags)
412{
413 struct rb_node *rb = object_tree(objflags)->rb_node;
414 unsigned long untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
415
416 while (rb) {
417 struct kmemleak_object *object;
418 unsigned long untagged_objp;
419
420 object = rb_entry(rb, struct kmemleak_object, rb_node);
421 untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
422
423 if (untagged_ptr < untagged_objp)
424 rb = object->rb_node.rb_left;
425 else if (untagged_objp + object->size <= untagged_ptr)
426 rb = object->rb_node.rb_right;
427 else if (untagged_objp == untagged_ptr || alias)
428 return object;
429 else {
430 kmemleak_warn("Found object by alias at 0x%08lx\n",
431 ptr);
432 dump_object_info(object);
433 break;
434 }
435 }
436 return NULL;
437}
438
439/* Look-up a kmemleak object which allocated with virtual address. */
440static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
441{
442 return __lookup_object(ptr, alias, 0);
443}
444
445/*
446 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
447 * that once an object's use_count reached 0, the RCU freeing was already
448 * registered and the object should no longer be used. This function must be
449 * called under the protection of rcu_read_lock().
450 */
451static int get_object(struct kmemleak_object *object)
452{
453 return atomic_inc_not_zero(&object->use_count);
454}
455
456/*
457 * Memory pool allocation and freeing. kmemleak_lock must not be held.
458 */
459static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
460{
461 unsigned long flags;
462 struct kmemleak_object *object;
463
464 /* try the slab allocator first */
465 if (object_cache) {
466 object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
467 if (object)
468 return object;
469 }
470
471 /* slab allocation failed, try the memory pool */
472 raw_spin_lock_irqsave(&kmemleak_lock, flags);
473 object = list_first_entry_or_null(&mem_pool_free_list,
474 typeof(*object), object_list);
475 if (object)
476 list_del(&object->object_list);
477 else if (mem_pool_free_count)
478 object = &mem_pool[--mem_pool_free_count];
479 else
480 pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
481 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
482
483 return object;
484}
485
486/*
487 * Return the object to either the slab allocator or the memory pool.
488 */
489static void mem_pool_free(struct kmemleak_object *object)
490{
491 unsigned long flags;
492
493 if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) {
494 kmem_cache_free(object_cache, object);
495 return;
496 }
497
498 /* add the object to the memory pool free list */
499 raw_spin_lock_irqsave(&kmemleak_lock, flags);
500 list_add(&object->object_list, &mem_pool_free_list);
501 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
502}
503
504/*
505 * RCU callback to free a kmemleak_object.
506 */
507static void free_object_rcu(struct rcu_head *rcu)
508{
509 struct hlist_node *tmp;
510 struct kmemleak_scan_area *area;
511 struct kmemleak_object *object =
512 container_of(rcu, struct kmemleak_object, rcu);
513
514 /*
515 * Once use_count is 0 (guaranteed by put_object), there is no other
516 * code accessing this object, hence no need for locking.
517 */
518 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
519 hlist_del(&area->node);
520 kmem_cache_free(scan_area_cache, area);
521 }
522 mem_pool_free(object);
523}
524
525/*
526 * Decrement the object use_count. Once the count is 0, free the object using
527 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
528 * delete_object() path, the delayed RCU freeing ensures that there is no
529 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
530 * is also possible.
531 */
532static void put_object(struct kmemleak_object *object)
533{
534 if (!atomic_dec_and_test(&object->use_count))
535 return;
536
537 /* should only get here after delete_object was called */
538 WARN_ON(object->flags & OBJECT_ALLOCATED);
539
540 /*
541 * It may be too early for the RCU callbacks, however, there is no
542 * concurrent object_list traversal when !object_cache and all objects
543 * came from the memory pool. Free the object directly.
544 */
545 if (object_cache)
546 call_rcu(&object->rcu, free_object_rcu);
547 else
548 free_object_rcu(&object->rcu);
549}
550
551/*
552 * Look up an object in the object search tree and increase its use_count.
553 */
554static struct kmemleak_object *__find_and_get_object(unsigned long ptr, int alias,
555 unsigned int objflags)
556{
557 unsigned long flags;
558 struct kmemleak_object *object;
559
560 rcu_read_lock();
561 raw_spin_lock_irqsave(&kmemleak_lock, flags);
562 object = __lookup_object(ptr, alias, objflags);
563 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
564
565 /* check whether the object is still available */
566 if (object && !get_object(object))
567 object = NULL;
568 rcu_read_unlock();
569
570 return object;
571}
572
573/* Look up and get an object which allocated with virtual address. */
574static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
575{
576 return __find_and_get_object(ptr, alias, 0);
577}
578
579/*
580 * Remove an object from its object tree and object_list. Must be called with
581 * the kmemleak_lock held _if_ kmemleak is still enabled.
582 */
583static void __remove_object(struct kmemleak_object *object)
584{
585 rb_erase(&object->rb_node, object_tree(object->flags));
586 if (!(object->del_state & DELSTATE_NO_DELETE))
587 list_del_rcu(&object->object_list);
588 object->del_state |= DELSTATE_REMOVED;
589}
590
591static struct kmemleak_object *__find_and_remove_object(unsigned long ptr,
592 int alias,
593 unsigned int objflags)
594{
595 struct kmemleak_object *object;
596
597 object = __lookup_object(ptr, alias, objflags);
598 if (object)
599 __remove_object(object);
600
601 return object;
602}
603
604/*
605 * Look up an object in the object search tree and remove it from both object
606 * tree root and object_list. The returned object's use_count should be at
607 * least 1, as initially set by create_object().
608 */
609static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias,
610 unsigned int objflags)
611{
612 unsigned long flags;
613 struct kmemleak_object *object;
614
615 raw_spin_lock_irqsave(&kmemleak_lock, flags);
616 object = __find_and_remove_object(ptr, alias, objflags);
617 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
618
619 return object;
620}
621
622static noinline depot_stack_handle_t set_track_prepare(void)
623{
624 depot_stack_handle_t trace_handle;
625 unsigned long entries[MAX_TRACE];
626 unsigned int nr_entries;
627
628 /*
629 * Use object_cache to determine whether kmemleak_init() has
630 * been invoked. stack_depot_early_init() is called before
631 * kmemleak_init() in mm_core_init().
632 */
633 if (!object_cache)
634 return 0;
635 nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);
636 trace_handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
637
638 return trace_handle;
639}
640
641static struct kmemleak_object *__alloc_object(gfp_t gfp)
642{
643 struct kmemleak_object *object;
644
645 object = mem_pool_alloc(gfp);
646 if (!object) {
647 pr_warn("Cannot allocate a kmemleak_object structure\n");
648 kmemleak_disable();
649 return NULL;
650 }
651
652 INIT_LIST_HEAD(&object->object_list);
653 INIT_LIST_HEAD(&object->gray_list);
654 INIT_HLIST_HEAD(&object->area_list);
655 raw_spin_lock_init(&object->lock);
656 atomic_set(&object->use_count, 1);
657 object->excess_ref = 0;
658 object->count = 0; /* white color initially */
659 object->checksum = 0;
660 object->del_state = 0;
661
662 /* task information */
663 if (in_hardirq()) {
664 object->pid = 0;
665 strncpy(object->comm, "hardirq", sizeof(object->comm));
666 } else if (in_serving_softirq()) {
667 object->pid = 0;
668 strncpy(object->comm, "softirq", sizeof(object->comm));
669 } else {
670 object->pid = current->pid;
671 /*
672 * There is a small chance of a race with set_task_comm(),
673 * however using get_task_comm() here may cause locking
674 * dependency issues with current->alloc_lock. In the worst
675 * case, the command line is not correct.
676 */
677 strncpy(object->comm, current->comm, sizeof(object->comm));
678 }
679
680 /* kernel backtrace */
681 object->trace_handle = set_track_prepare();
682
683 return object;
684}
685
686static int __link_object(struct kmemleak_object *object, unsigned long ptr,
687 size_t size, int min_count, unsigned int objflags)
688{
689
690 struct kmemleak_object *parent;
691 struct rb_node **link, *rb_parent;
692 unsigned long untagged_ptr;
693 unsigned long untagged_objp;
694
695 object->flags = OBJECT_ALLOCATED | objflags;
696 object->pointer = ptr;
697 object->size = kfence_ksize((void *)ptr) ?: size;
698 object->min_count = min_count;
699 object->jiffies = jiffies;
700
701 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
702 /*
703 * Only update min_addr and max_addr with object
704 * storing virtual address.
705 */
706 if (!(objflags & (OBJECT_PHYS | OBJECT_PERCPU))) {
707 min_addr = min(min_addr, untagged_ptr);
708 max_addr = max(max_addr, untagged_ptr + size);
709 }
710 link = &object_tree(objflags)->rb_node;
711 rb_parent = NULL;
712 while (*link) {
713 rb_parent = *link;
714 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
715 untagged_objp = (unsigned long)kasan_reset_tag((void *)parent->pointer);
716 if (untagged_ptr + size <= untagged_objp)
717 link = &parent->rb_node.rb_left;
718 else if (untagged_objp + parent->size <= untagged_ptr)
719 link = &parent->rb_node.rb_right;
720 else {
721 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
722 ptr);
723 /*
724 * No need for parent->lock here since "parent" cannot
725 * be freed while the kmemleak_lock is held.
726 */
727 dump_object_info(parent);
728 return -EEXIST;
729 }
730 }
731 rb_link_node(&object->rb_node, rb_parent, link);
732 rb_insert_color(&object->rb_node, object_tree(objflags));
733 list_add_tail_rcu(&object->object_list, &object_list);
734
735 return 0;
736}
737
738/*
739 * Create the metadata (struct kmemleak_object) corresponding to an allocated
740 * memory block and add it to the object_list and object tree.
741 */
742static void __create_object(unsigned long ptr, size_t size,
743 int min_count, gfp_t gfp, unsigned int objflags)
744{
745 struct kmemleak_object *object;
746 unsigned long flags;
747 int ret;
748
749 object = __alloc_object(gfp);
750 if (!object)
751 return;
752
753 raw_spin_lock_irqsave(&kmemleak_lock, flags);
754 ret = __link_object(object, ptr, size, min_count, objflags);
755 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
756 if (ret)
757 mem_pool_free(object);
758}
759
760/* Create kmemleak object which allocated with virtual address. */
761static void create_object(unsigned long ptr, size_t size,
762 int min_count, gfp_t gfp)
763{
764 __create_object(ptr, size, min_count, gfp, 0);
765}
766
767/* Create kmemleak object which allocated with physical address. */
768static void create_object_phys(unsigned long ptr, size_t size,
769 int min_count, gfp_t gfp)
770{
771 __create_object(ptr, size, min_count, gfp, OBJECT_PHYS);
772}
773
774/* Create kmemleak object corresponding to a per-CPU allocation. */
775static void create_object_percpu(unsigned long ptr, size_t size,
776 int min_count, gfp_t gfp)
777{
778 __create_object(ptr, size, min_count, gfp, OBJECT_PERCPU);
779}
780
781/*
782 * Mark the object as not allocated and schedule RCU freeing via put_object().
783 */
784static void __delete_object(struct kmemleak_object *object)
785{
786 unsigned long flags;
787
788 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
789 WARN_ON(atomic_read(&object->use_count) < 1);
790
791 /*
792 * Locking here also ensures that the corresponding memory block
793 * cannot be freed when it is being scanned.
794 */
795 raw_spin_lock_irqsave(&object->lock, flags);
796 object->flags &= ~OBJECT_ALLOCATED;
797 raw_spin_unlock_irqrestore(&object->lock, flags);
798 put_object(object);
799}
800
801/*
802 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
803 * delete it.
804 */
805static void delete_object_full(unsigned long ptr, unsigned int objflags)
806{
807 struct kmemleak_object *object;
808
809 object = find_and_remove_object(ptr, 0, objflags);
810 if (!object) {
811#ifdef DEBUG
812 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
813 ptr);
814#endif
815 return;
816 }
817 __delete_object(object);
818}
819
820/*
821 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
822 * delete it. If the memory block is partially freed, the function may create
823 * additional metadata for the remaining parts of the block.
824 */
825static void delete_object_part(unsigned long ptr, size_t size,
826 unsigned int objflags)
827{
828 struct kmemleak_object *object, *object_l, *object_r;
829 unsigned long start, end, flags;
830
831 object_l = __alloc_object(GFP_KERNEL);
832 if (!object_l)
833 return;
834
835 object_r = __alloc_object(GFP_KERNEL);
836 if (!object_r)
837 goto out;
838
839 raw_spin_lock_irqsave(&kmemleak_lock, flags);
840 object = __find_and_remove_object(ptr, 1, objflags);
841 if (!object) {
842#ifdef DEBUG
843 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
844 ptr, size);
845#endif
846 goto unlock;
847 }
848
849 /*
850 * Create one or two objects that may result from the memory block
851 * split. Note that partial freeing is only done by free_bootmem() and
852 * this happens before kmemleak_init() is called.
853 */
854 start = object->pointer;
855 end = object->pointer + object->size;
856 if ((ptr > start) &&
857 !__link_object(object_l, start, ptr - start,
858 object->min_count, objflags))
859 object_l = NULL;
860 if ((ptr + size < end) &&
861 !__link_object(object_r, ptr + size, end - ptr - size,
862 object->min_count, objflags))
863 object_r = NULL;
864
865unlock:
866 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
867 if (object)
868 __delete_object(object);
869
870out:
871 if (object_l)
872 mem_pool_free(object_l);
873 if (object_r)
874 mem_pool_free(object_r);
875}
876
877static void __paint_it(struct kmemleak_object *object, int color)
878{
879 object->min_count = color;
880 if (color == KMEMLEAK_BLACK)
881 object->flags |= OBJECT_NO_SCAN;
882}
883
884static void paint_it(struct kmemleak_object *object, int color)
885{
886 unsigned long flags;
887
888 raw_spin_lock_irqsave(&object->lock, flags);
889 __paint_it(object, color);
890 raw_spin_unlock_irqrestore(&object->lock, flags);
891}
892
893static void paint_ptr(unsigned long ptr, int color, unsigned int objflags)
894{
895 struct kmemleak_object *object;
896
897 object = __find_and_get_object(ptr, 0, objflags);
898 if (!object) {
899 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
900 ptr,
901 (color == KMEMLEAK_GREY) ? "Grey" :
902 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
903 return;
904 }
905 paint_it(object, color);
906 put_object(object);
907}
908
909/*
910 * Mark an object permanently as gray-colored so that it can no longer be
911 * reported as a leak. This is used in general to mark a false positive.
912 */
913static void make_gray_object(unsigned long ptr)
914{
915 paint_ptr(ptr, KMEMLEAK_GREY, 0);
916}
917
918/*
919 * Mark the object as black-colored so that it is ignored from scans and
920 * reporting.
921 */
922static void make_black_object(unsigned long ptr, unsigned int objflags)
923{
924 paint_ptr(ptr, KMEMLEAK_BLACK, objflags);
925}
926
927/*
928 * Add a scanning area to the object. If at least one such area is added,
929 * kmemleak will only scan these ranges rather than the whole memory block.
930 */
931static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
932{
933 unsigned long flags;
934 struct kmemleak_object *object;
935 struct kmemleak_scan_area *area = NULL;
936 unsigned long untagged_ptr;
937 unsigned long untagged_objp;
938
939 object = find_and_get_object(ptr, 1);
940 if (!object) {
941 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
942 ptr);
943 return;
944 }
945
946 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
947 untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
948
949 if (scan_area_cache)
950 area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
951
952 raw_spin_lock_irqsave(&object->lock, flags);
953 if (!area) {
954 pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
955 /* mark the object for full scan to avoid false positives */
956 object->flags |= OBJECT_FULL_SCAN;
957 goto out_unlock;
958 }
959 if (size == SIZE_MAX) {
960 size = untagged_objp + object->size - untagged_ptr;
961 } else if (untagged_ptr + size > untagged_objp + object->size) {
962 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
963 dump_object_info(object);
964 kmem_cache_free(scan_area_cache, area);
965 goto out_unlock;
966 }
967
968 INIT_HLIST_NODE(&area->node);
969 area->start = ptr;
970 area->size = size;
971
972 hlist_add_head(&area->node, &object->area_list);
973out_unlock:
974 raw_spin_unlock_irqrestore(&object->lock, flags);
975 put_object(object);
976}
977
978/*
979 * Any surplus references (object already gray) to 'ptr' are passed to
980 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
981 * vm_struct may be used as an alternative reference to the vmalloc'ed object
982 * (see free_thread_stack()).
983 */
984static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
985{
986 unsigned long flags;
987 struct kmemleak_object *object;
988
989 object = find_and_get_object(ptr, 0);
990 if (!object) {
991 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
992 ptr);
993 return;
994 }
995
996 raw_spin_lock_irqsave(&object->lock, flags);
997 object->excess_ref = excess_ref;
998 raw_spin_unlock_irqrestore(&object->lock, flags);
999 put_object(object);
1000}
1001
1002/*
1003 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
1004 * pointer. Such object will not be scanned by kmemleak but references to it
1005 * are searched.
1006 */
1007static void object_no_scan(unsigned long ptr)
1008{
1009 unsigned long flags;
1010 struct kmemleak_object *object;
1011
1012 object = find_and_get_object(ptr, 0);
1013 if (!object) {
1014 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
1015 return;
1016 }
1017
1018 raw_spin_lock_irqsave(&object->lock, flags);
1019 object->flags |= OBJECT_NO_SCAN;
1020 raw_spin_unlock_irqrestore(&object->lock, flags);
1021 put_object(object);
1022}
1023
1024/**
1025 * kmemleak_alloc - register a newly allocated object
1026 * @ptr: pointer to beginning of the object
1027 * @size: size of the object
1028 * @min_count: minimum number of references to this object. If during memory
1029 * scanning a number of references less than @min_count is found,
1030 * the object is reported as a memory leak. If @min_count is 0,
1031 * the object is never reported as a leak. If @min_count is -1,
1032 * the object is ignored (not scanned and not reported as a leak)
1033 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1034 *
1035 * This function is called from the kernel allocators when a new object
1036 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
1037 */
1038void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
1039 gfp_t gfp)
1040{
1041 pr_debug("%s(0x%px, %zu, %d)\n", __func__, ptr, size, min_count);
1042
1043 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1044 create_object((unsigned long)ptr, size, min_count, gfp);
1045}
1046EXPORT_SYMBOL_GPL(kmemleak_alloc);
1047
1048/**
1049 * kmemleak_alloc_percpu - register a newly allocated __percpu object
1050 * @ptr: __percpu pointer to beginning of the object
1051 * @size: size of the object
1052 * @gfp: flags used for kmemleak internal memory allocations
1053 *
1054 * This function is called from the kernel percpu allocator when a new object
1055 * (memory block) is allocated (alloc_percpu).
1056 */
1057void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
1058 gfp_t gfp)
1059{
1060 pr_debug("%s(0x%px, %zu)\n", __func__, ptr, size);
1061
1062 /*
1063 * Percpu allocations are only scanned and not reported as leaks
1064 * (min_count is set to 0).
1065 */
1066 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1067 create_object_percpu((unsigned long)ptr, size, 0, gfp);
1068}
1069EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
1070
1071/**
1072 * kmemleak_vmalloc - register a newly vmalloc'ed object
1073 * @area: pointer to vm_struct
1074 * @size: size of the object
1075 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
1076 *
1077 * This function is called from the vmalloc() kernel allocator when a new
1078 * object (memory block) is allocated.
1079 */
1080void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
1081{
1082 pr_debug("%s(0x%px, %zu)\n", __func__, area, size);
1083
1084 /*
1085 * A min_count = 2 is needed because vm_struct contains a reference to
1086 * the virtual address of the vmalloc'ed block.
1087 */
1088 if (kmemleak_enabled) {
1089 create_object((unsigned long)area->addr, size, 2, gfp);
1090 object_set_excess_ref((unsigned long)area,
1091 (unsigned long)area->addr);
1092 }
1093}
1094EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
1095
1096/**
1097 * kmemleak_free - unregister a previously registered object
1098 * @ptr: pointer to beginning of the object
1099 *
1100 * This function is called from the kernel allocators when an object (memory
1101 * block) is freed (kmem_cache_free, kfree, vfree etc.).
1102 */
1103void __ref kmemleak_free(const void *ptr)
1104{
1105 pr_debug("%s(0x%px)\n", __func__, ptr);
1106
1107 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1108 delete_object_full((unsigned long)ptr, 0);
1109}
1110EXPORT_SYMBOL_GPL(kmemleak_free);
1111
1112/**
1113 * kmemleak_free_part - partially unregister a previously registered object
1114 * @ptr: pointer to the beginning or inside the object. This also
1115 * represents the start of the range to be freed
1116 * @size: size to be unregistered
1117 *
1118 * This function is called when only a part of a memory block is freed
1119 * (usually from the bootmem allocator).
1120 */
1121void __ref kmemleak_free_part(const void *ptr, size_t size)
1122{
1123 pr_debug("%s(0x%px)\n", __func__, ptr);
1124
1125 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1126 delete_object_part((unsigned long)ptr, size, 0);
1127}
1128EXPORT_SYMBOL_GPL(kmemleak_free_part);
1129
1130/**
1131 * kmemleak_free_percpu - unregister a previously registered __percpu object
1132 * @ptr: __percpu pointer to beginning of the object
1133 *
1134 * This function is called from the kernel percpu allocator when an object
1135 * (memory block) is freed (free_percpu).
1136 */
1137void __ref kmemleak_free_percpu(const void __percpu *ptr)
1138{
1139 pr_debug("%s(0x%px)\n", __func__, ptr);
1140
1141 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1142 delete_object_full((unsigned long)ptr, OBJECT_PERCPU);
1143}
1144EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1145
1146/**
1147 * kmemleak_update_trace - update object allocation stack trace
1148 * @ptr: pointer to beginning of the object
1149 *
1150 * Override the object allocation stack trace for cases where the actual
1151 * allocation place is not always useful.
1152 */
1153void __ref kmemleak_update_trace(const void *ptr)
1154{
1155 struct kmemleak_object *object;
1156 depot_stack_handle_t trace_handle;
1157 unsigned long flags;
1158
1159 pr_debug("%s(0x%px)\n", __func__, ptr);
1160
1161 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1162 return;
1163
1164 object = find_and_get_object((unsigned long)ptr, 1);
1165 if (!object) {
1166#ifdef DEBUG
1167 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1168 ptr);
1169#endif
1170 return;
1171 }
1172
1173 trace_handle = set_track_prepare();
1174 raw_spin_lock_irqsave(&object->lock, flags);
1175 object->trace_handle = trace_handle;
1176 raw_spin_unlock_irqrestore(&object->lock, flags);
1177
1178 put_object(object);
1179}
1180EXPORT_SYMBOL(kmemleak_update_trace);
1181
1182/**
1183 * kmemleak_not_leak - mark an allocated object as false positive
1184 * @ptr: pointer to beginning of the object
1185 *
1186 * Calling this function on an object will cause the memory block to no longer
1187 * be reported as leak and always be scanned.
1188 */
1189void __ref kmemleak_not_leak(const void *ptr)
1190{
1191 pr_debug("%s(0x%px)\n", __func__, ptr);
1192
1193 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1194 make_gray_object((unsigned long)ptr);
1195}
1196EXPORT_SYMBOL(kmemleak_not_leak);
1197
1198/**
1199 * kmemleak_ignore - ignore an allocated object
1200 * @ptr: pointer to beginning of the object
1201 *
1202 * Calling this function on an object will cause the memory block to be
1203 * ignored (not scanned and not reported as a leak). This is usually done when
1204 * it is known that the corresponding block is not a leak and does not contain
1205 * any references to other allocated memory blocks.
1206 */
1207void __ref kmemleak_ignore(const void *ptr)
1208{
1209 pr_debug("%s(0x%px)\n", __func__, ptr);
1210
1211 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1212 make_black_object((unsigned long)ptr, 0);
1213}
1214EXPORT_SYMBOL(kmemleak_ignore);
1215
1216/**
1217 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1218 * @ptr: pointer to beginning or inside the object. This also
1219 * represents the start of the scan area
1220 * @size: size of the scan area
1221 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1222 *
1223 * This function is used when it is known that only certain parts of an object
1224 * contain references to other objects. Kmemleak will only scan these areas
1225 * reducing the number false negatives.
1226 */
1227void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1228{
1229 pr_debug("%s(0x%px)\n", __func__, ptr);
1230
1231 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1232 add_scan_area((unsigned long)ptr, size, gfp);
1233}
1234EXPORT_SYMBOL(kmemleak_scan_area);
1235
1236/**
1237 * kmemleak_no_scan - do not scan an allocated object
1238 * @ptr: pointer to beginning of the object
1239 *
1240 * This function notifies kmemleak not to scan the given memory block. Useful
1241 * in situations where it is known that the given object does not contain any
1242 * references to other objects. Kmemleak will not scan such objects reducing
1243 * the number of false negatives.
1244 */
1245void __ref kmemleak_no_scan(const void *ptr)
1246{
1247 pr_debug("%s(0x%px)\n", __func__, ptr);
1248
1249 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1250 object_no_scan((unsigned long)ptr);
1251}
1252EXPORT_SYMBOL(kmemleak_no_scan);
1253
1254/**
1255 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1256 * address argument
1257 * @phys: physical address of the object
1258 * @size: size of the object
1259 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1260 */
1261void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, gfp_t gfp)
1262{
1263 pr_debug("%s(0x%px, %zu)\n", __func__, &phys, size);
1264
1265 if (kmemleak_enabled)
1266 /*
1267 * Create object with OBJECT_PHYS flag and
1268 * assume min_count 0.
1269 */
1270 create_object_phys((unsigned long)phys, size, 0, gfp);
1271}
1272EXPORT_SYMBOL(kmemleak_alloc_phys);
1273
1274/**
1275 * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1276 * physical address argument
1277 * @phys: physical address if the beginning or inside an object. This
1278 * also represents the start of the range to be freed
1279 * @size: size to be unregistered
1280 */
1281void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1282{
1283 pr_debug("%s(0x%px)\n", __func__, &phys);
1284
1285 if (kmemleak_enabled)
1286 delete_object_part((unsigned long)phys, size, OBJECT_PHYS);
1287}
1288EXPORT_SYMBOL(kmemleak_free_part_phys);
1289
1290/**
1291 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1292 * address argument
1293 * @phys: physical address of the object
1294 */
1295void __ref kmemleak_ignore_phys(phys_addr_t phys)
1296{
1297 pr_debug("%s(0x%px)\n", __func__, &phys);
1298
1299 if (kmemleak_enabled)
1300 make_black_object((unsigned long)phys, OBJECT_PHYS);
1301}
1302EXPORT_SYMBOL(kmemleak_ignore_phys);
1303
1304/*
1305 * Update an object's checksum and return true if it was modified.
1306 */
1307static bool update_checksum(struct kmemleak_object *object)
1308{
1309 u32 old_csum = object->checksum;
1310
1311 if (WARN_ON_ONCE(object->flags & (OBJECT_PHYS | OBJECT_PERCPU)))
1312 return false;
1313
1314 kasan_disable_current();
1315 kcsan_disable_current();
1316 object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size);
1317 kasan_enable_current();
1318 kcsan_enable_current();
1319
1320 return object->checksum != old_csum;
1321}
1322
1323/*
1324 * Update an object's references. object->lock must be held by the caller.
1325 */
1326static void update_refs(struct kmemleak_object *object)
1327{
1328 if (!color_white(object)) {
1329 /* non-orphan, ignored or new */
1330 return;
1331 }
1332
1333 /*
1334 * Increase the object's reference count (number of pointers to the
1335 * memory block). If this count reaches the required minimum, the
1336 * object's color will become gray and it will be added to the
1337 * gray_list.
1338 */
1339 object->count++;
1340 if (color_gray(object)) {
1341 /* put_object() called when removing from gray_list */
1342 WARN_ON(!get_object(object));
1343 list_add_tail(&object->gray_list, &gray_list);
1344 }
1345}
1346
1347/*
1348 * Memory scanning is a long process and it needs to be interruptible. This
1349 * function checks whether such interrupt condition occurred.
1350 */
1351static int scan_should_stop(void)
1352{
1353 if (!kmemleak_enabled)
1354 return 1;
1355
1356 /*
1357 * This function may be called from either process or kthread context,
1358 * hence the need to check for both stop conditions.
1359 */
1360 if (current->mm)
1361 return signal_pending(current);
1362 else
1363 return kthread_should_stop();
1364
1365 return 0;
1366}
1367
1368/*
1369 * Scan a memory block (exclusive range) for valid pointers and add those
1370 * found to the gray list.
1371 */
1372static void scan_block(void *_start, void *_end,
1373 struct kmemleak_object *scanned)
1374{
1375 unsigned long *ptr;
1376 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1377 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1378 unsigned long flags;
1379 unsigned long untagged_ptr;
1380
1381 raw_spin_lock_irqsave(&kmemleak_lock, flags);
1382 for (ptr = start; ptr < end; ptr++) {
1383 struct kmemleak_object *object;
1384 unsigned long pointer;
1385 unsigned long excess_ref;
1386
1387 if (scan_should_stop())
1388 break;
1389
1390 kasan_disable_current();
1391 pointer = *(unsigned long *)kasan_reset_tag((void *)ptr);
1392 kasan_enable_current();
1393
1394 untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
1395 if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
1396 continue;
1397
1398 /*
1399 * No need for get_object() here since we hold kmemleak_lock.
1400 * object->use_count cannot be dropped to 0 while the object
1401 * is still present in object_tree_root and object_list
1402 * (with updates protected by kmemleak_lock).
1403 */
1404 object = lookup_object(pointer, 1);
1405 if (!object)
1406 continue;
1407 if (object == scanned)
1408 /* self referenced, ignore */
1409 continue;
1410
1411 /*
1412 * Avoid the lockdep recursive warning on object->lock being
1413 * previously acquired in scan_object(). These locks are
1414 * enclosed by scan_mutex.
1415 */
1416 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1417 /* only pass surplus references (object already gray) */
1418 if (color_gray(object)) {
1419 excess_ref = object->excess_ref;
1420 /* no need for update_refs() if object already gray */
1421 } else {
1422 excess_ref = 0;
1423 update_refs(object);
1424 }
1425 raw_spin_unlock(&object->lock);
1426
1427 if (excess_ref) {
1428 object = lookup_object(excess_ref, 0);
1429 if (!object)
1430 continue;
1431 if (object == scanned)
1432 /* circular reference, ignore */
1433 continue;
1434 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1435 update_refs(object);
1436 raw_spin_unlock(&object->lock);
1437 }
1438 }
1439 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
1440}
1441
1442/*
1443 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1444 */
1445#ifdef CONFIG_SMP
1446static void scan_large_block(void *start, void *end)
1447{
1448 void *next;
1449
1450 while (start < end) {
1451 next = min(start + MAX_SCAN_SIZE, end);
1452 scan_block(start, next, NULL);
1453 start = next;
1454 cond_resched();
1455 }
1456}
1457#endif
1458
1459/*
1460 * Scan a memory block corresponding to a kmemleak_object. A condition is
1461 * that object->use_count >= 1.
1462 */
1463static void scan_object(struct kmemleak_object *object)
1464{
1465 struct kmemleak_scan_area *area;
1466 unsigned long flags;
1467
1468 /*
1469 * Once the object->lock is acquired, the corresponding memory block
1470 * cannot be freed (the same lock is acquired in delete_object).
1471 */
1472 raw_spin_lock_irqsave(&object->lock, flags);
1473 if (object->flags & OBJECT_NO_SCAN)
1474 goto out;
1475 if (!(object->flags & OBJECT_ALLOCATED))
1476 /* already freed object */
1477 goto out;
1478
1479 if (object->flags & OBJECT_PERCPU) {
1480 unsigned int cpu;
1481
1482 for_each_possible_cpu(cpu) {
1483 void *start = per_cpu_ptr((void __percpu *)object->pointer, cpu);
1484 void *end = start + object->size;
1485
1486 scan_block(start, end, object);
1487
1488 raw_spin_unlock_irqrestore(&object->lock, flags);
1489 cond_resched();
1490 raw_spin_lock_irqsave(&object->lock, flags);
1491 if (!(object->flags & OBJECT_ALLOCATED))
1492 break;
1493 }
1494 } else if (hlist_empty(&object->area_list) ||
1495 object->flags & OBJECT_FULL_SCAN) {
1496 void *start = object->flags & OBJECT_PHYS ?
1497 __va((phys_addr_t)object->pointer) :
1498 (void *)object->pointer;
1499 void *end = start + object->size;
1500 void *next;
1501
1502 do {
1503 next = min(start + MAX_SCAN_SIZE, end);
1504 scan_block(start, next, object);
1505
1506 start = next;
1507 if (start >= end)
1508 break;
1509
1510 raw_spin_unlock_irqrestore(&object->lock, flags);
1511 cond_resched();
1512 raw_spin_lock_irqsave(&object->lock, flags);
1513 } while (object->flags & OBJECT_ALLOCATED);
1514 } else {
1515 hlist_for_each_entry(area, &object->area_list, node)
1516 scan_block((void *)area->start,
1517 (void *)(area->start + area->size),
1518 object);
1519 }
1520out:
1521 raw_spin_unlock_irqrestore(&object->lock, flags);
1522}
1523
1524/*
1525 * Scan the objects already referenced (gray objects). More objects will be
1526 * referenced and, if there are no memory leaks, all the objects are scanned.
1527 */
1528static void scan_gray_list(void)
1529{
1530 struct kmemleak_object *object, *tmp;
1531
1532 /*
1533 * The list traversal is safe for both tail additions and removals
1534 * from inside the loop. The kmemleak objects cannot be freed from
1535 * outside the loop because their use_count was incremented.
1536 */
1537 object = list_entry(gray_list.next, typeof(*object), gray_list);
1538 while (&object->gray_list != &gray_list) {
1539 cond_resched();
1540
1541 /* may add new objects to the list */
1542 if (!scan_should_stop())
1543 scan_object(object);
1544
1545 tmp = list_entry(object->gray_list.next, typeof(*object),
1546 gray_list);
1547
1548 /* remove the object from the list and release it */
1549 list_del(&object->gray_list);
1550 put_object(object);
1551
1552 object = tmp;
1553 }
1554 WARN_ON(!list_empty(&gray_list));
1555}
1556
1557/*
1558 * Conditionally call resched() in an object iteration loop while making sure
1559 * that the given object won't go away without RCU read lock by performing a
1560 * get_object() if necessaary.
1561 */
1562static void kmemleak_cond_resched(struct kmemleak_object *object)
1563{
1564 if (!get_object(object))
1565 return; /* Try next object */
1566
1567 raw_spin_lock_irq(&kmemleak_lock);
1568 if (object->del_state & DELSTATE_REMOVED)
1569 goto unlock_put; /* Object removed */
1570 object->del_state |= DELSTATE_NO_DELETE;
1571 raw_spin_unlock_irq(&kmemleak_lock);
1572
1573 rcu_read_unlock();
1574 cond_resched();
1575 rcu_read_lock();
1576
1577 raw_spin_lock_irq(&kmemleak_lock);
1578 if (object->del_state & DELSTATE_REMOVED)
1579 list_del_rcu(&object->object_list);
1580 object->del_state &= ~DELSTATE_NO_DELETE;
1581unlock_put:
1582 raw_spin_unlock_irq(&kmemleak_lock);
1583 put_object(object);
1584}
1585
1586/*
1587 * Scan data sections and all the referenced memory blocks allocated via the
1588 * kernel's standard allocators. This function must be called with the
1589 * scan_mutex held.
1590 */
1591static void kmemleak_scan(void)
1592{
1593 struct kmemleak_object *object;
1594 struct zone *zone;
1595 int __maybe_unused i;
1596 int new_leaks = 0;
1597
1598 jiffies_last_scan = jiffies;
1599
1600 /* prepare the kmemleak_object's */
1601 rcu_read_lock();
1602 list_for_each_entry_rcu(object, &object_list, object_list) {
1603 raw_spin_lock_irq(&object->lock);
1604#ifdef DEBUG
1605 /*
1606 * With a few exceptions there should be a maximum of
1607 * 1 reference to any object at this point.
1608 */
1609 if (atomic_read(&object->use_count) > 1) {
1610 pr_debug("object->use_count = %d\n",
1611 atomic_read(&object->use_count));
1612 dump_object_info(object);
1613 }
1614#endif
1615
1616 /* ignore objects outside lowmem (paint them black) */
1617 if ((object->flags & OBJECT_PHYS) &&
1618 !(object->flags & OBJECT_NO_SCAN)) {
1619 unsigned long phys = object->pointer;
1620
1621 if (PHYS_PFN(phys) < min_low_pfn ||
1622 PHYS_PFN(phys + object->size) >= max_low_pfn)
1623 __paint_it(object, KMEMLEAK_BLACK);
1624 }
1625
1626 /* reset the reference count (whiten the object) */
1627 object->count = 0;
1628 if (color_gray(object) && get_object(object))
1629 list_add_tail(&object->gray_list, &gray_list);
1630
1631 raw_spin_unlock_irq(&object->lock);
1632
1633 if (need_resched())
1634 kmemleak_cond_resched(object);
1635 }
1636 rcu_read_unlock();
1637
1638#ifdef CONFIG_SMP
1639 /* per-cpu sections scanning */
1640 for_each_possible_cpu(i)
1641 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1642 __per_cpu_end + per_cpu_offset(i));
1643#endif
1644
1645 /*
1646 * Struct page scanning for each node.
1647 */
1648 get_online_mems();
1649 for_each_populated_zone(zone) {
1650 unsigned long start_pfn = zone->zone_start_pfn;
1651 unsigned long end_pfn = zone_end_pfn(zone);
1652 unsigned long pfn;
1653
1654 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1655 struct page *page = pfn_to_online_page(pfn);
1656
1657 if (!(pfn & 63))
1658 cond_resched();
1659
1660 if (!page)
1661 continue;
1662
1663 /* only scan pages belonging to this zone */
1664 if (page_zone(page) != zone)
1665 continue;
1666 /* only scan if page is in use */
1667 if (page_count(page) == 0)
1668 continue;
1669 scan_block(page, page + 1, NULL);
1670 }
1671 }
1672 put_online_mems();
1673
1674 /*
1675 * Scanning the task stacks (may introduce false negatives).
1676 */
1677 if (kmemleak_stack_scan) {
1678 struct task_struct *p, *g;
1679
1680 rcu_read_lock();
1681 for_each_process_thread(g, p) {
1682 void *stack = try_get_task_stack(p);
1683 if (stack) {
1684 scan_block(stack, stack + THREAD_SIZE, NULL);
1685 put_task_stack(p);
1686 }
1687 }
1688 rcu_read_unlock();
1689 }
1690
1691 /*
1692 * Scan the objects already referenced from the sections scanned
1693 * above.
1694 */
1695 scan_gray_list();
1696
1697 /*
1698 * Check for new or unreferenced objects modified since the previous
1699 * scan and color them gray until the next scan.
1700 */
1701 rcu_read_lock();
1702 list_for_each_entry_rcu(object, &object_list, object_list) {
1703 if (need_resched())
1704 kmemleak_cond_resched(object);
1705
1706 /*
1707 * This is racy but we can save the overhead of lock/unlock
1708 * calls. The missed objects, if any, should be caught in
1709 * the next scan.
1710 */
1711 if (!color_white(object))
1712 continue;
1713 raw_spin_lock_irq(&object->lock);
1714 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1715 && update_checksum(object) && get_object(object)) {
1716 /* color it gray temporarily */
1717 object->count = object->min_count;
1718 list_add_tail(&object->gray_list, &gray_list);
1719 }
1720 raw_spin_unlock_irq(&object->lock);
1721 }
1722 rcu_read_unlock();
1723
1724 /*
1725 * Re-scan the gray list for modified unreferenced objects.
1726 */
1727 scan_gray_list();
1728
1729 /*
1730 * If scanning was stopped do not report any new unreferenced objects.
1731 */
1732 if (scan_should_stop())
1733 return;
1734
1735 /*
1736 * Scanning result reporting.
1737 */
1738 rcu_read_lock();
1739 list_for_each_entry_rcu(object, &object_list, object_list) {
1740 if (need_resched())
1741 kmemleak_cond_resched(object);
1742
1743 /*
1744 * This is racy but we can save the overhead of lock/unlock
1745 * calls. The missed objects, if any, should be caught in
1746 * the next scan.
1747 */
1748 if (!color_white(object))
1749 continue;
1750 raw_spin_lock_irq(&object->lock);
1751 if (unreferenced_object(object) &&
1752 !(object->flags & OBJECT_REPORTED)) {
1753 object->flags |= OBJECT_REPORTED;
1754
1755 if (kmemleak_verbose)
1756 print_unreferenced(NULL, object);
1757
1758 new_leaks++;
1759 }
1760 raw_spin_unlock_irq(&object->lock);
1761 }
1762 rcu_read_unlock();
1763
1764 if (new_leaks) {
1765 kmemleak_found_leaks = true;
1766
1767 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1768 new_leaks);
1769 }
1770
1771}
1772
1773/*
1774 * Thread function performing automatic memory scanning. Unreferenced objects
1775 * at the end of a memory scan are reported but only the first time.
1776 */
1777static int kmemleak_scan_thread(void *arg)
1778{
1779 static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
1780
1781 pr_info("Automatic memory scanning thread started\n");
1782 set_user_nice(current, 10);
1783
1784 /*
1785 * Wait before the first scan to allow the system to fully initialize.
1786 */
1787 if (first_run) {
1788 signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
1789 first_run = 0;
1790 while (timeout && !kthread_should_stop())
1791 timeout = schedule_timeout_interruptible(timeout);
1792 }
1793
1794 while (!kthread_should_stop()) {
1795 signed long timeout = READ_ONCE(jiffies_scan_wait);
1796
1797 mutex_lock(&scan_mutex);
1798 kmemleak_scan();
1799 mutex_unlock(&scan_mutex);
1800
1801 /* wait before the next scan */
1802 while (timeout && !kthread_should_stop())
1803 timeout = schedule_timeout_interruptible(timeout);
1804 }
1805
1806 pr_info("Automatic memory scanning thread ended\n");
1807
1808 return 0;
1809}
1810
1811/*
1812 * Start the automatic memory scanning thread. This function must be called
1813 * with the scan_mutex held.
1814 */
1815static void start_scan_thread(void)
1816{
1817 if (scan_thread)
1818 return;
1819 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1820 if (IS_ERR(scan_thread)) {
1821 pr_warn("Failed to create the scan thread\n");
1822 scan_thread = NULL;
1823 }
1824}
1825
1826/*
1827 * Stop the automatic memory scanning thread.
1828 */
1829static void stop_scan_thread(void)
1830{
1831 if (scan_thread) {
1832 kthread_stop(scan_thread);
1833 scan_thread = NULL;
1834 }
1835}
1836
1837/*
1838 * Iterate over the object_list and return the first valid object at or after
1839 * the required position with its use_count incremented. The function triggers
1840 * a memory scanning when the pos argument points to the first position.
1841 */
1842static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1843{
1844 struct kmemleak_object *object;
1845 loff_t n = *pos;
1846 int err;
1847
1848 err = mutex_lock_interruptible(&scan_mutex);
1849 if (err < 0)
1850 return ERR_PTR(err);
1851
1852 rcu_read_lock();
1853 list_for_each_entry_rcu(object, &object_list, object_list) {
1854 if (n-- > 0)
1855 continue;
1856 if (get_object(object))
1857 goto out;
1858 }
1859 object = NULL;
1860out:
1861 return object;
1862}
1863
1864/*
1865 * Return the next object in the object_list. The function decrements the
1866 * use_count of the previous object and increases that of the next one.
1867 */
1868static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1869{
1870 struct kmemleak_object *prev_obj = v;
1871 struct kmemleak_object *next_obj = NULL;
1872 struct kmemleak_object *obj = prev_obj;
1873
1874 ++(*pos);
1875
1876 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1877 if (get_object(obj)) {
1878 next_obj = obj;
1879 break;
1880 }
1881 }
1882
1883 put_object(prev_obj);
1884 return next_obj;
1885}
1886
1887/*
1888 * Decrement the use_count of the last object required, if any.
1889 */
1890static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1891{
1892 if (!IS_ERR(v)) {
1893 /*
1894 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1895 * waiting was interrupted, so only release it if !IS_ERR.
1896 */
1897 rcu_read_unlock();
1898 mutex_unlock(&scan_mutex);
1899 if (v)
1900 put_object(v);
1901 }
1902}
1903
1904/*
1905 * Print the information for an unreferenced object to the seq file.
1906 */
1907static int kmemleak_seq_show(struct seq_file *seq, void *v)
1908{
1909 struct kmemleak_object *object = v;
1910 unsigned long flags;
1911
1912 raw_spin_lock_irqsave(&object->lock, flags);
1913 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1914 print_unreferenced(seq, object);
1915 raw_spin_unlock_irqrestore(&object->lock, flags);
1916 return 0;
1917}
1918
1919static const struct seq_operations kmemleak_seq_ops = {
1920 .start = kmemleak_seq_start,
1921 .next = kmemleak_seq_next,
1922 .stop = kmemleak_seq_stop,
1923 .show = kmemleak_seq_show,
1924};
1925
1926static int kmemleak_open(struct inode *inode, struct file *file)
1927{
1928 return seq_open(file, &kmemleak_seq_ops);
1929}
1930
1931static int dump_str_object_info(const char *str)
1932{
1933 unsigned long flags;
1934 struct kmemleak_object *object;
1935 unsigned long addr;
1936
1937 if (kstrtoul(str, 0, &addr))
1938 return -EINVAL;
1939 object = find_and_get_object(addr, 0);
1940 if (!object) {
1941 pr_info("Unknown object at 0x%08lx\n", addr);
1942 return -EINVAL;
1943 }
1944
1945 raw_spin_lock_irqsave(&object->lock, flags);
1946 dump_object_info(object);
1947 raw_spin_unlock_irqrestore(&object->lock, flags);
1948
1949 put_object(object);
1950 return 0;
1951}
1952
1953/*
1954 * We use grey instead of black to ensure we can do future scans on the same
1955 * objects. If we did not do future scans these black objects could
1956 * potentially contain references to newly allocated objects in the future and
1957 * we'd end up with false positives.
1958 */
1959static void kmemleak_clear(void)
1960{
1961 struct kmemleak_object *object;
1962
1963 rcu_read_lock();
1964 list_for_each_entry_rcu(object, &object_list, object_list) {
1965 raw_spin_lock_irq(&object->lock);
1966 if ((object->flags & OBJECT_REPORTED) &&
1967 unreferenced_object(object))
1968 __paint_it(object, KMEMLEAK_GREY);
1969 raw_spin_unlock_irq(&object->lock);
1970 }
1971 rcu_read_unlock();
1972
1973 kmemleak_found_leaks = false;
1974}
1975
1976static void __kmemleak_do_cleanup(void);
1977
1978/*
1979 * File write operation to configure kmemleak at run-time. The following
1980 * commands can be written to the /sys/kernel/debug/kmemleak file:
1981 * off - disable kmemleak (irreversible)
1982 * stack=on - enable the task stacks scanning
1983 * stack=off - disable the tasks stacks scanning
1984 * scan=on - start the automatic memory scanning thread
1985 * scan=off - stop the automatic memory scanning thread
1986 * scan=... - set the automatic memory scanning period in seconds (0 to
1987 * disable it)
1988 * scan - trigger a memory scan
1989 * clear - mark all current reported unreferenced kmemleak objects as
1990 * grey to ignore printing them, or free all kmemleak objects
1991 * if kmemleak has been disabled.
1992 * dump=... - dump information about the object found at the given address
1993 */
1994static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1995 size_t size, loff_t *ppos)
1996{
1997 char buf[64];
1998 int buf_size;
1999 int ret;
2000
2001 buf_size = min(size, (sizeof(buf) - 1));
2002 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
2003 return -EFAULT;
2004 buf[buf_size] = 0;
2005
2006 ret = mutex_lock_interruptible(&scan_mutex);
2007 if (ret < 0)
2008 return ret;
2009
2010 if (strncmp(buf, "clear", 5) == 0) {
2011 if (kmemleak_enabled)
2012 kmemleak_clear();
2013 else
2014 __kmemleak_do_cleanup();
2015 goto out;
2016 }
2017
2018 if (!kmemleak_enabled) {
2019 ret = -EPERM;
2020 goto out;
2021 }
2022
2023 if (strncmp(buf, "off", 3) == 0)
2024 kmemleak_disable();
2025 else if (strncmp(buf, "stack=on", 8) == 0)
2026 kmemleak_stack_scan = 1;
2027 else if (strncmp(buf, "stack=off", 9) == 0)
2028 kmemleak_stack_scan = 0;
2029 else if (strncmp(buf, "scan=on", 7) == 0)
2030 start_scan_thread();
2031 else if (strncmp(buf, "scan=off", 8) == 0)
2032 stop_scan_thread();
2033 else if (strncmp(buf, "scan=", 5) == 0) {
2034 unsigned secs;
2035 unsigned long msecs;
2036
2037 ret = kstrtouint(buf + 5, 0, &secs);
2038 if (ret < 0)
2039 goto out;
2040
2041 msecs = secs * MSEC_PER_SEC;
2042 if (msecs > UINT_MAX)
2043 msecs = UINT_MAX;
2044
2045 stop_scan_thread();
2046 if (msecs) {
2047 WRITE_ONCE(jiffies_scan_wait, msecs_to_jiffies(msecs));
2048 start_scan_thread();
2049 }
2050 } else if (strncmp(buf, "scan", 4) == 0)
2051 kmemleak_scan();
2052 else if (strncmp(buf, "dump=", 5) == 0)
2053 ret = dump_str_object_info(buf + 5);
2054 else
2055 ret = -EINVAL;
2056
2057out:
2058 mutex_unlock(&scan_mutex);
2059 if (ret < 0)
2060 return ret;
2061
2062 /* ignore the rest of the buffer, only one command at a time */
2063 *ppos += size;
2064 return size;
2065}
2066
2067static const struct file_operations kmemleak_fops = {
2068 .owner = THIS_MODULE,
2069 .open = kmemleak_open,
2070 .read = seq_read,
2071 .write = kmemleak_write,
2072 .llseek = seq_lseek,
2073 .release = seq_release,
2074};
2075
2076static void __kmemleak_do_cleanup(void)
2077{
2078 struct kmemleak_object *object, *tmp;
2079
2080 /*
2081 * Kmemleak has already been disabled, no need for RCU list traversal
2082 * or kmemleak_lock held.
2083 */
2084 list_for_each_entry_safe(object, tmp, &object_list, object_list) {
2085 __remove_object(object);
2086 __delete_object(object);
2087 }
2088}
2089
2090/*
2091 * Stop the memory scanning thread and free the kmemleak internal objects if
2092 * no previous scan thread (otherwise, kmemleak may still have some useful
2093 * information on memory leaks).
2094 */
2095static void kmemleak_do_cleanup(struct work_struct *work)
2096{
2097 stop_scan_thread();
2098
2099 mutex_lock(&scan_mutex);
2100 /*
2101 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
2102 * longer track object freeing. Ordering of the scan thread stopping and
2103 * the memory accesses below is guaranteed by the kthread_stop()
2104 * function.
2105 */
2106 kmemleak_free_enabled = 0;
2107 mutex_unlock(&scan_mutex);
2108
2109 if (!kmemleak_found_leaks)
2110 __kmemleak_do_cleanup();
2111 else
2112 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
2113}
2114
2115static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
2116
2117/*
2118 * Disable kmemleak. No memory allocation/freeing will be traced once this
2119 * function is called. Disabling kmemleak is an irreversible operation.
2120 */
2121static void kmemleak_disable(void)
2122{
2123 /* atomically check whether it was already invoked */
2124 if (cmpxchg(&kmemleak_error, 0, 1))
2125 return;
2126
2127 /* stop any memory operation tracing */
2128 kmemleak_enabled = 0;
2129
2130 /* check whether it is too early for a kernel thread */
2131 if (kmemleak_late_initialized)
2132 schedule_work(&cleanup_work);
2133 else
2134 kmemleak_free_enabled = 0;
2135
2136 pr_info("Kernel memory leak detector disabled\n");
2137}
2138
2139/*
2140 * Allow boot-time kmemleak disabling (enabled by default).
2141 */
2142static int __init kmemleak_boot_config(char *str)
2143{
2144 if (!str)
2145 return -EINVAL;
2146 if (strcmp(str, "off") == 0)
2147 kmemleak_disable();
2148 else if (strcmp(str, "on") == 0) {
2149 kmemleak_skip_disable = 1;
2150 stack_depot_request_early_init();
2151 }
2152 else
2153 return -EINVAL;
2154 return 0;
2155}
2156early_param("kmemleak", kmemleak_boot_config);
2157
2158/*
2159 * Kmemleak initialization.
2160 */
2161void __init kmemleak_init(void)
2162{
2163#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
2164 if (!kmemleak_skip_disable) {
2165 kmemleak_disable();
2166 return;
2167 }
2168#endif
2169
2170 if (kmemleak_error)
2171 return;
2172
2173 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
2174 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
2175
2176 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
2177 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
2178
2179 /* register the data/bss sections */
2180 create_object((unsigned long)_sdata, _edata - _sdata,
2181 KMEMLEAK_GREY, GFP_ATOMIC);
2182 create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
2183 KMEMLEAK_GREY, GFP_ATOMIC);
2184 /* only register .data..ro_after_init if not within .data */
2185 if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata)
2186 create_object((unsigned long)__start_ro_after_init,
2187 __end_ro_after_init - __start_ro_after_init,
2188 KMEMLEAK_GREY, GFP_ATOMIC);
2189}
2190
2191/*
2192 * Late initialization function.
2193 */
2194static int __init kmemleak_late_init(void)
2195{
2196 kmemleak_late_initialized = 1;
2197
2198 debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops);
2199
2200 if (kmemleak_error) {
2201 /*
2202 * Some error occurred and kmemleak was disabled. There is a
2203 * small chance that kmemleak_disable() was called immediately
2204 * after setting kmemleak_late_initialized and we may end up with
2205 * two clean-up threads but serialized by scan_mutex.
2206 */
2207 schedule_work(&cleanup_work);
2208 return -ENOMEM;
2209 }
2210
2211 if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
2212 mutex_lock(&scan_mutex);
2213 start_scan_thread();
2214 mutex_unlock(&scan_mutex);
2215 }
2216
2217 pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
2218 mem_pool_free_count);
2219
2220 return 0;
2221}
2222late_initcall(kmemleak_late_init);